KASLR• iOS 6 introduces KASLR CanSecWest 2013 - iOS 6 Exploitation 280 Days Later

KASLR• iOS 6 introduces KASLR

1. http://www.sektioneins.de iOS 6 Exploitation 280 Days LaterStefan Esser <stefan.esser@sektioneins.de> CanSecWest Vancouver
2. Who am I? Stefan Esser • from Cologne / Germany • in information security since 1998 • initially did a lot of low level security • from 2001 to 2010 focused on PHP / web app security • since mid-2010 focused on iPhone security (ASLR, kernel exploitation) • Head of Research and Development at SektionEins GmbH Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  2
3. What is this talk about?• iOS 6 is the new major version of iOS with tons of new security features• new kernel security mitigations already discussed by Mark Dowd/Tarjei Mandt• but iOS 6.x has other not yet mentioned new security features• and some kernel features require commentary• basically an update to my CSW 2012 talk• 280 days later because it was about 280 days later when I submitted to Dragos Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  3
4. Part IiOS Security Timeline 2012-2013 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  4
5. CanSecWest 2012 - iOS 5 An Exploitation Nightmare?March 2012• reasons why iOS 5 jailbreak took so long• history of some iOS security features• history of iOS security bugfixes• getting kernel debugger running on new devices• abusing BPF as kernel weird machineURL: http://cansecwest.com/csw12/CSW2012_StefanEsser_iOS5_An_Exploitation_Nightmare_FINAL.pdf Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  5
6. iOS Hacker‘s HandbookApril 2012• Charlie Miller - Dionysius Blazakis - Dino Dai Zovi• Stefan Esser - Vincenzo Iozzo - Ralf-Philipp Weinmann• covers iOS 4 to iOS 5• iOS Security Basics, iOS in the Enterprise• Encryption, Code Signing and Memory Protection• Sandboxing, Fuzzing iOS Applications• Exploitation, Return-Oriented-Programming• Kernel-Debugging and Exploitation, Jailbreaking, Baseband AttacksURL: http://ca.wiley.com/WileyCDA/WileyTitle/productCd-1118204123.html Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  6
7. SyScan 2012 - iOS Kernel Heap ArmageddonApril 2012• different iOS kernel heap wrappers• feasibility of cross zone / memory manager attacks• attacking IOKit application data / object vtables instead of heap meta data• using OSUnserializeXML() for generic kernel level heap feng shui• talk updated for BlackHat USA & XCon 2012URL 1: http://reverse.put.as/wp-content/uploads/2011/06/SyScan2012_StefanEsser_iOS_Kernel_Heap_Armageddon.pdfURL 2: http://media.blackhat.com/bh-us-12/Briefings/Esser/BH_US_12_Esser_iOS_Kernel_Heap_Armageddon_WP.pdf Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  7
8. FinFisher Mobile - The Smartphone Who Loved MeAugust 2012• by CitizenLab• analysis of FinFisher for mobile devices• samples caught in the wild• iOS sample compiled for developer phones• media wrongly assumed developer cert lets you write spy applicationsURL: https://citizenlab.org/2012/08/the-smartphone-who-loved-me-finfisher-goes-mobile/ Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  8
9. FinSpy Moile: iOS and Apple UDID LeakSeptember 2012• by Alex Radocea^Crowdstrike• deep analysis of FinFisher for iOS• revealed that there was no iOS priv escape 0-day in FinFisher iOS - just empty placeholder• instead seems to heavily rely on being jailbroken with a public jailbreak prior to installationURL: http://www.crowdstrike.com/blog/finspy-mobile-ios-and-apple-udid-leak/index.html Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  9
10. iOS 6 Released and J/“F“ailbroken on Day 1September 2012 • by Musclenerd • iOS 6 on pre-A5 already tethered jailbroken on day one • by CHPWN • iOS 6 on iPhone 5 already failbroken on day one • failbroken means Cydia runs but no kernel payloadURL: https://twitter.com/chpwn/status/249249908094296064 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  10
11. HITB2012 - iOS 6 Kernel SecurityOctober 2012 • by Mark Dowd and Tarjei Mandt • deep analysis of new iOS 6 kernel exploit mitigations • contained a 0-day kernel info leak vulnerability • and the vm_map_copy exploitation technique heavily used by latest iOS 6 jailbreakURL: http://conference.hackinthebox.org/hitbsecconf2012kul/materials/D1T2%20-%20Mark%20Dowd%20&%20Tarjei%20Mandt%20-%20iOS6%20Security.pdfVideo: http://www.youtube.com/watch?v=O-WZinEoki4 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  11
12. POC2012 - Find your own iOS kernel bug November 2012• by Xu Hao and Chen Xiaobo• analysis of previous IOKit vulnerability• about fuzzing iOKit for vulnerabilities• later repeated at SyScan360 in DecemberURL: http://syscan.org/index.php/download/get/328bf4b37e6ae8b799472ff230465339/XuHao_Chen_Xiaobo_Find_your_own_iOS_kernel_bug.zip Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  12
13. Hackulo.us / Installous shutdown December 2012 • announcement that Hackulo.us shut down • also took down Installous the notorious application used by iOS application pirates on jailbroken iPhones • celebrated by media, jailbreak developers and iOS app developers around the worldURL: http://thanks-god-not-anymo.re Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  13
14. kuaiyong, Zeusmos, 25pp, ... January 2013 • after installous is dead more and more iOS piracy solutions that do not require jailbreak • solutions reportedly based on account sharing and/or some undisclosed exploit • still active ?!?URL 1: http://m.csoonline.com/article/725183/now-pirated-ios-apps-can-be-installed-without-jailbreakURL 2: http://no.you.dont.get.the.url.you.want Research Assistant: Marc Rogers Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  14
15. Community Milking and iOS 6 JB Release February 2013• by evad3rs• website with donation button and multiple banner ads• told people repeatedly for about a week to check website for status updates• about one week later release of iOS 6.0/6.1 jailbreak• so far the most expensive jailbreak in terms of crowdfundingURL: http://www.evasi0n.com/ Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  15
16. evasi0n Jailbreak‘s Userland Components February 2013 • by Braden Thomas^AccuvantLabs • analysis of userland components of evasi0n jailbreak • covers most of the userland bugs exploited by evasi0nURL: http://blog.accuvantlabs.com/blog/bthomas/evasi0n-jailbreaks-userland-component Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  16
17. Dissecting the “evasi0n“ Kernel Exploit February 2013 • by Tarjei Mandt^Azimuth • analysis of kernel components of evasi0n jailbreak • shows how evasi0n is based on techniques discussed in the iOS 6 kernel security talk by azimuthURL: http://blog.azimuthsecurity.com/2013/02/from-usr-to-svc-dissecting-evasi0n.html Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  17
18. Part IIiOS 6 Kernel Security “Improvements“ Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  18
19. KASLR• iOS 6 introduces KASLR - kernel address space layout randomization• only 256 possible load addresses• each 2 MB apart• starting at 0x81200000 ending at 0xA1000000 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  19
20. KASLR: But why 2 MB Aligned? • 2 MB alignment of KASLR seems arbitrary • why not smaller alignment? • big alignment is less secure • right now: • leak any address in __DATA and you know the kernel‘s base address (address - 0x200000) & 0xFFE00000 • leak any address from first 2 MB of kernel __TEXT and know the kernel‘s base address address & 0xFFE00000 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  20
21. Kernel Address Space Hardening• kernel __TEXT no longer writable ➡ to stop kernel code hotpatching• kernel heap no longer executable ➡ to stop just executing kernel data• kernel address space is separated from user space processes ➡ to stop return into user space code and offset from NULL-deref attacks Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  21
22. Kernel Stack Cookies • iOS 6 added stack cookies to protect from kernel stack buffer overflows • implementation is rather unusual • stack cookie on top of stack • bottom of local stack contains ptr to the value it is compared against • second byte of stack cookie is forced to 0x00 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  22
23. Kernel Stack Cookie Verification• stack cookie verification in function epilog• verification against cookie pointed to• fact that stack_cookie_ptr and stack_cookie are both on stack is a weakness• wrong cookie value will lead to a kernel panic without message Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  23
24. Kernel Heap Cookies• iOS 4 and iOS 5 kernel heap exploitation has always attacked the free list• in iOS 6 Apple introduced heap protection cookies to protect free list• distinguishes between small poisoned and larger non-poisoned blocks• two different security cookies are used for this➡ stops attacks against the free list as used before in public jailbreaks Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  24
25. Kernel Heap Cookies (larger blocks)• for larger blocks the memory content is kept but end is trashed with cookie• secret cookie has lowest bit cleared• if data of freed block leaks this leaks • a kernel heap address: 0x87b46500 • the secret cookie: 0x6b7769c8 ^ 0x87b46500 = 0xECC30CC8 next_pointer 87b46480: 00 65 b4 87 00 00 00 00 00 00 00 00 00 00 00 00 .e.............. 87b46490: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464b0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464c0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464d0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464e0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 87b464f0: 00 00 00 00 00 00 00 00 00 00 00 00 c8 69 77 6b .............iwk next_pointer^non_poisoned_cookie Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  25
26. Kernel Heap Cookies (small blocks)• for small blocks the memory content is overwritten with 0xdeadbeef• secret cookie has lowest bit set• if data of freed block leaks this leaks • a kernel heap address: 0x92f1c740 • the secret cookie: 0x7ec1387b ^ 0x92f1c740 = 0xEC30FF3B next_pointer 92f1c700: 40 c7 f1 92 ef be ad de ef be ad de ef be ad de @............... 92f1c710: ef be ad de ef be ad de ef be ad de ef be ad de ................ 92f1c720: ef be ad de ef be ad de ef be ad de ef be ad de ................ 92f1c730: ef be ad de ef be ad de ef be ad de 7b 38 c1 7e ............{8.~ next_pointer^poisoned_cookie Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  26
27. Kernel Heap Cookies after allocation• on allocation free list pointer and cookie are overwritten with 0xdeadbeef• most probably as defense in depth against information leaks 9072b000: ef be ad de 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b010: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b020: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b030: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b040: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b050: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b060: 00 00 00 ff 00 00 00 ff 00 00 00 ff 00 00 00 ff ................ 9072b070: 00 00 00 ff 00 00 00 ff 00 00 00 ff ef be ad de ................ Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  27
28. Kernel Heap Hardening• previously mach_zone_info() and host_zone_info() leaked internal state• both functions now require debugging kernel boot arguments• previously OSUnserializeXML() allowed fine control over kernel heap• Apple fixed some bugs in it and put some arbitrary limits on it• only exact methods described at BlackHat / SyScan were killed• other ways to abuse this function for kernel heap feng shui still working Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  28
29. Death to Kernel Info Leaks• two fold strategy to fight kernel info leaks • fix information leak vulnerabilities • obfuscate kernel addresses returned to user land• example of fixed information leaks • BPF stack data info leak • kern.proc leak fixed • kern.file info leak fixed Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  29
30. Kernel Address Obfuscation• lots of kernel API return kernel addresses to user land processes e.g. mach_port_kobject(), mach_port_space_info(), vm_region_recurse(), vm_map_region_recurse(), vm_map_page_info(), proc_info(), fstat(), sysctl()• protected by adding a random 32 bit cookie (lowest bit set) Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  30
31. Kernel Image Address Obfuscation• some API might even return addresses inside the kernel image• these addresses are additionally unslid to protect against KASLR leaks Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  31
32. Readonly Syscall Table• previous jailbreaks used partial syscall table overwrites• Apple moved syscall table into section __DATA::__const• section is made read only at runtime• controlled by kernel boot argument dataconstro• stops syscall table corruption ... Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  32
33. Just replace Syscall Table completely?• kernel linking changes in iOS 6 introduced lots of indirect accesses• syscall table is no longer accessed directly (also true for lots of other stuff)• instead pointer to syscall table is used from __nl_symbol_ptr section• and guess what - this section is writable Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  33
34. Part IIIiOS 6 Misc Hardening Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  34
35. BPF not so weird anymore... • at CSW 2012 BPF was mentioned as weird machine inside the kernel • in iOS 6.x it is still a machine but not so weird anymore • Apple added sanity checks inside the function • access to slack memory is now checked for bounds Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  35
36. mobile_house_arrest - Readonly Code Directory • lockdown service for reading / writing into app directories • since iOS 6 application‘s code directory is no longer writable • previously it was possible to replace arbitrary application resources Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  36
37. Part IVUser Space ASLR (Address Space Layout Randomization) Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  37
38. ASLR in iOS 4.3-6.x • randomly slides • main binary • dyld (dynamic linker) • dynamic library cache Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  38
39. Position Independent Executables in 2012 $ python ipapiescan.py Adobe Reader - armv7 - PIE - N/A Bluefire Reader - armv6|armv7 - NO_PIE - 3.0 DiamondDash - armv7 - NO_PIE - 4.2 Ebook Reader - armv6|armv7 - NO_PIE - N/A eBookS Reader - armv6|armv7 - NO_PIE - N/A Facebook - armv6|armv7 - NO_PIE - 4.0 Fly With Me - armv6|armv7 - NO_PIE - 3.0 FPK Reader - armv6|armv7 - NO_PIE - 3.2 Hotels - armv6|armv7 - NO_PIE - 3.1 iBooks - armv6|armv7 - NO_PIE - 4.2 • all system binaries were KakaoTalk - armv6|armv7 - NO_PIE - 3.1 Messenger - armv6|armv7 - NO_PIE - 4.0 compiled as PIE PerfectReader Mini - armv6|armv7 - NO_PIE - N/A QR Reader - armv6|armv7 - NO_PIE - 4.0 QR Scanner QR-Scanner - - armv6|armv7 armv7 - - NO_PIE NO_PIE - - N/A 4.0 • most 3rd party apps were QRCode Quick Scan - - armv6|armv7 armv6|armv7 - - NO_PIE NO_PIE - - N/A 4.0 not compiled as PIE Skype - armv6|armv7 - NO_PIE - N/A Twitter - armv6|armv7 - NO_PIE - 4.0 vBookz PDF - armv7 - PIE - 4.3 VZ-Netzwerke - armv6 - NO_PIE - 3.0 Wallpapers - armv6|armv7 - NO_PIE - 4.1 WhatsApp - armv6|armv7 - NO_PIE - 3.1 Where is - armv6|armv7 - NO_PIE - 4.1source code of old idapiescan.py is available at Githubhttps://github.com/stefanesser/idapiescan Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  39
40. iOS 4.3-6.x: NO PIE main binary randomization • dynamic loader is not slid in iOS 4 for NO PIE main executables • since iOS 5 the dynamic loader is always slid • randomized by kernel in 256 positions iOS 4.3 - 4.3.x - NO PIE main executable iOS 5.0 - 6.x - NO PIE main executable Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  40
41. Position Independent Executables in 2013 $ python ipapiescan.py Bluefire Reader - armv7(s) - PIE - 4.3 Calendar Pro - armv7(s) - PIE - 4.3 CalenMob - armv7(s) - PIE - 5.0 Chrome - armv7 - PIE - 4.3 CloudOn - armv7 - NO_PIE - 5.0 DiamondDash - armv7(s) - PIE - 4.3 Documents - armv7(s) - PIE - 4.3 • all system binaries are Ebook Reader - armv7 - PIE - 4.3 eBookS Reader - armv6|armv7 - NO_PIE - N/A compiled as PIE Facebook - armv7 - PIE - 4.3 G-Whizz! - armv6|armv7 - NO_PIE - 4.0 Gmail Google - - armv7 armv7 - - PIE PIE - - 5.0 4.3 • most 3rd party apps are Google Drive - armv7 - PIE - 5.0 now compiled as PIE Google Earth - armv7 - PIE - 4.3 Google+ - armv7 - PIE - 5.0 iBooks IM+ - - armv7 armv7(s) - - PIE PIE - - 5.0 4.3 • NO_PIE mostly Instagram - armv7 - PIE - 4.3 KakaoTalk - armv7(s) - PIE - 4.3 unimportant apps Latitude - armv6|armv7 - NO_PIE - N/A Local - armv6|armv7 - PIE - 4.3 Lync 2010 Messenger - - armv7 armv7 - - NO_PIE PIE - - 4.3 4.3 • some high profile MSN World - armv7(s) - PIE - 4.3 exceptions are: Skype, SkyDrive - armv6|armv7 - NO_PIE - 4.0 Skype - armv7 - NO_PIE - 4.3 SkyDrive, SmartGlass - armv7 - PIE - 5.0 SSH Mobile Free - armv7(s) - PIE - 4.3 Google Translate, ... SystemTools - armv7(s) - PIE - 4.3 Translate - armv6|armv7 - NO_PIE - N/A Trillian - armv7 - PIE - 4.3 Twitter - armv7 - PIE - 5.0 Usessh - armv7(s) - PIE - 4.3 Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  41
42. iOS 4.3-6.0: PIE main binary randomization • for PIE main executables the main binary and dyld are randomized • main binary and dyld are slid the same amount • randomized by kernel in 256 positions iOS 4.3 - 6.0 - PIE main executable Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  42
43. iOS 6.1: PIE main binary randomization • since iOS 6.1 the kernel finally generates two separate slides • randomness of both is still limited to 256 positions • knowing addresses in dyld / main no longer leaks address of other iOS 4.3 - 6.0 - PIE main executable iOS 6.1 - PIE main executable Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  43
44. iOS 4.3-5.1.1: dyld_shared_cache randomization • data and code must slide together (due to codesigning) • hole after code - data usually loaded to 0x3E000000 • max slide determined by difference of end of shared area and end of data • around 4200 different positions iOS 4.3 - 5.1.1 - no slide iOS 4.3 - 5.1.1 - maximum slide Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  44
45. iOS 6.x: dyld_shared_cache randomization • code and data loaded right next to each other • no more hole - no more wasted space • max slide determined by size of shared area minus size of shared cache • about 21500 different positions for iPod 4G (new devices = more code = less random) iOS 6.x - no slide iOS 6.x - maximum slide Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  45
46. Part ViOS 6 and the Partial Code-signing Vulnerability Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  46
47. Partial Code-signing Vulnerability (iOS 4) • in iOS 4.x jailbreaks the method of choice to launch untether exploits • when a mach-o is loaded the kernel will load it as is • a possible signature will be registered • missing signature is okay until a not signed executable page is accessed • dyld was tricked with malformed mach-o data structures to execute code Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  47
48. sniffLoadCommands (iOS 4.3.4) • function does pre-handling of mach-o load commands • iOS 4.3.4 adds protection against partial code signing • mach-o load commands must be inside a segment • mach-o load commands can only be in R + X segment • mach-o load commands may not be partially in a segment ➡ effect is that once dyld maps the header R+X it can only continue to work on it if there is a valid signature Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  48
49. Partial Code-signing Vuln (iOS 4.3.4-iOS 5.1.1) • protection in sniffLoadCommands could be bypassed • by having a RW- LC_SEGMENT64 for mach-o header • and a fake R-X LC_SEGMENT for mach-o header • disclosed at CanSecWest 2012 - here on stage • worked because kernel handles LC_SEGMENT64 and dyld did not • magic is that dyld will read mach-o header from from address in memory Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  49
50. sniffLoadCommands (iOS 6.0) • iOS 6.0 adds protection against CSW 2012 trick to sniffLoadCommands • if a LC_SEGMENT64 load command is found an exception is thrown ➡ CSW 2012 trick was already partially broken after iOS 5.1.1 • in iOS 5.1.1 AMFI verified existence of a code signing blob Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  50
51. Load Command Segment Override (iOS 6.0-6.1.2) • bug used by evasi0n • kernel not directly involved in loading dynamic libraries only dyld is • dyld could be tricked with a malicious dylib • contains real R-X segment with load commands in it • contains second R-- segment that contains copy of load commands • virtual address of both segments is set to same position • later segment in mach-o will replace previous in memory • when dyld accesses header it is in RO memory = no sig needed = bypass Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  51
52. sniffLoadCommands (iOS 6.1.3 beta 2) • iOS 6.1.3 beta 2 adds additional protections to sniffLoadCommands • load commands must now be in one segment only • for dynamic libraries a second sniff pass is added • all segments must not intersect the R-X segment containing the load commands ➡ evasi0n untether killed Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  52
53. Part VIiOS 6.1 and Launch-Daemon-Code-Signing Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  53
54. Launch Daemons to launch Untethers • in iOS 5.x jailbreaks were launched on boot via launch daemons • launch daemons are plists describing system services <?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http:// www.apple.com/DTDs/PropertyList-1.0.dtd"> <plist version="1.0"> <dict> <key>Label</key> <string>jb</string> <key>ProgramArguments</key> <array> <string>/usr/sbin/corona</string> <string>-f</string> <string>racoon-exploit.conf</string> </array> <key>WorkingDirectory</key> <string>/usr/share/corona/</string> <key>RunAtLoad</key> <true/> <key>LaunchOnlyOnce</key> <true/> <key>DisableAslr</key> DisableAslr was removed from iOS 5.1 <true/> </dict> </plist> Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  54
55. Launch-Daemon-Code-Signing (I)• abuse of launch daemons lead to new iOS 6.1 security feature• launch daemon loading is now code signed• implemented in /bin/launchctl• can be bypassed by setting kernel boot arguments (not possible without low-level exploit) Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  55
56. Launch-Daemon-Code-Signing (II)• without launch-daemon-code-signing scans /System/Library/LaunchDaemons for defined /bin/launchctl launch daemons and loads them• with activated launch-daemon-code-signing a big plist with all defined launch daemons is loaded instead• launch daemon can only be loaded if it is defined in the plist and exists on disk Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  56
57. Launch-Daemon-Code-Signing (III)• big launch daemon plist is loaded from /System/Library/Caches/com.apple.xpcd/xpcd_cache.dylib• this dynamic library is within the dyld_shared_cache and therefore code signed• symbol __xpcd_cache must exist• but binary plist is take from sectiondata of __TEXT::__xpcd_cache Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  57
58. XPCD_CACHE.PLIST Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  58
59. Launch-Daemon-Code-Signing Security How secure Apple wanted Launch-Daemon-Code-Signing to be... Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  59
60. Launch-Daemon-Code-Signing Security How secure Launch-Daemon-Code-Signing is right now... Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  60
61. Launch-Daemon-Code-Signing Security • code signing itself seems to stop loading arbitrary launch daemons • but Apple forgot / or ignored /etc/launchd.conf • /etc/launchd.conf defines commands launchctl executes on start • attacker can execute arbitrary existing commands bsexec .. /sbin/mount -u -o rw,suid,dev / setenv DYLD_INSERT_LIBRARIES /private/var/evasi0n/amfi.dylib load /System/Library/LaunchDaemons/com.apple.MobileFileIntegrity.plist bsexec .. /private/var/evasi0n/evasi0n unsetenv DYLD_INSERT_LIBRARIES bsexec .. /bin/rm -f /private/var/evasi0n/sock bsexec .. /bin/ln -f /var/tmp/launchd/sock /private/var/evasi0n/sock Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  61
62. FAQ: Why not put old launchctl binary on device? Q: “If only the newest iOS 6.1 launchctl binary implements this code signing. Why not put an iOS 6.0 launchctl binary on the device to bypass this protection?“ A: “System binaries like launchctl do not come with a valid code signing signature from Apple. Instead they come only with the table of memory page hashes and entitlements. When the kernel loads such a binary it hashes these tables and checks if the hash is in a whitelist inside the kernel (a.k.a. trust cache). The old launchctl binary will not be accepted because it is not in the trust cache of the new kernel.“ Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  62
63. Final Words • with iOS 6 Apple has tried to kill all public techniques • finally kills some stuff that was previously known and ignored for 10 years • the new mitigations make exploitation a lot harder • when launch daemon code signing is hardened a bit more, persisting on iDevices will become incredibly hard • however there are still weaknesses in most of the protections • ... and tons of kernel information leaks Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  63
64. Questions ? Stefan Esser • iOS 6 - Exploitation 280 Days Later • March 2013 •  64