tpwn Privilege Escalation Vulnerability
Just days after Apple patched the DYLD_PRINT_TO_FILE security hole with the release of OS X 10.10.5, a developer has found a similar unpatched exploit that could allow attackers to gain root-level access to a Mac.
Luca Todesco shared information (via AppleInsider) on the “tpwn” exploit on GitHub over the weekend. It affects all versions of OS X Yosemite, including OS X 10.10.5, but does not affect OS X El Capitan.
1) I cannot really discuss specifics, but this particular bug would have been hard to find via a traditional IOKit fuzz, since it requires an invalid ‘task’ port passed over to IOServiceOpen. Most fuzzers use mach_task_self for that, and fuzz method calls/traps/properties/etc.
2) When IOServiceRelease is called, vtable+0x20 is called. the vtable pointer is controlled, at +0x20 I place a stack pivot, which sets RSP = RAX and pops 3 times. At 0x18 I place a POP RAX;RET gadget to let the chain begin after 0x28. Payload then locates the credentials structure, sets UID to 0 by bzero()ing, cleans up the memory corruption, decreases the task count for current user and increases task count for root. It then unlocks locks held by IOAudioEngine to prevent your audio from freezing up, and then returns to the userland context.
There is no weakness in address randomization I relied on for exploitation.
It relies on two distinct bugs, an info-leak to obtain a pointer to an allocation in the kalloc.1024 zone and a memory corruption primitive (deriving from a NULL pointer dfr. in IOKit) allowing me to OR 0x10 anywhere in kernel memory.
To break kASLR I corrupt the size of a vm_map_copy struct, which allows me to read the adjacent allocation to the struct, which is a C++ object. First 8 bytes of said C++ object is a pointer to the vtable, which resides in __TEXT of some kernel extension. Since I can calculate the unslid address from userland without any issue, by subtracting what gets leaked with what gets calculated you get to know the kASLR slide.
Just to clarify: The code execution part has 100% reliability rate. The kASLR leaking part does have some chance in it, however empirical evidence indicates that the failure rate is extremely low.
Todesco, who said he does security research in his spare time, said he notified Apple of the problems “a few hours before the exploit was published.”
“This is not due to me having issues with Apple’s patch policies/time frames, as others have incorrectly reported,” he wrote.
He also developed a patch called NULLGuard, which he’s included in the GitHub material. Since he does not have a Mac developer certificate, he wrote that he can’t distribute an easy-to-install version of the patch.
Update (2015-09-02): Luca Todesco:
So, tpwn gains root on any OS X box running a system <= 10.10.5 by gaining knowledge of the kernel address space layout randomisation slide followed by kernel code execution kickstarting a stack pivot which allows me to control the stack pointer of a kernel thread belonging to a controlled task.
Stefan Esser of DYLD exploit fame has added a patch for tpwn to his signed SUIDGuard kext (originally created for the DYLD exploit) to protect against tpwn. https://www.suidguard.com/stories/index.html