macOS Thread Injection via Task port
Last updated
Last updated
Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE) Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)
Initially, the task_threads() function is invoked on the task port to obtain a thread list from the remote task. A thread is selected for hijacking. This approach diverges from conventional code injection methods as creating a new remote thread is prohibited due to the new mitigation blocking thread_create_running().
To control the thread, thread_suspend() is called, halting its execution.
The only operations permitted on the remote thread involve stopping and starting it, retrieving and modifying its register values. Remote function calls are initiated by setting registers x0 to x7 to the arguments, configuring pc to target the desired function, and activating the thread. Ensuring the thread does not crash after the return necessitates detection of the return.
One strategy involves registering an exception handler for the remote thread using thread_set_exception_ports(), setting the lr register to an invalid address before the function call. This triggers an exception post-function execution, sending a message to the exception port, enabling state inspection of the thread to recover the return value. Alternatively, as adopted from Ian Beer’s triple_fetch exploit, lr is set to loop infinitely. The thread's registers are then continuously monitored until pc points to that instruction.
The subsequent phase involves establishing Mach ports to facilitate communication with the remote thread. These ports are instrumental in transferring arbitrary send and receive rights between tasks.
For bidirectional communication, two Mach receive rights are created: one in the local and the other in the remote task. Subsequently, a send right for each port is transferred to the counterpart task, enabling message exchange.
Focusing on the local port, the receive right is held by the local task. The port is created with mach_port_allocate(). The challenge lies in transferring a send right to this port into the remote task.
A strategy involves leveraging thread_set_special_port() to place a send right to the local port in the remote thread’s THREAD_KERNEL_PORT. Then, the remote thread is instructed to call mach_thread_self() to retrieve the send right.
For the remote port, the process is essentially reversed. The remote thread is directed to generate a Mach port via mach_reply_port() (as mach_port_allocate() is unsuitable due to its return mechanism). Upon port creation, mach_port_insert_right() is invoked in the remote thread to establish a send right. This right is then stashed in the kernel using thread_set_special_port(). Back in the local task, thread_get_special_port() is used on the remote thread to acquire a send right to the newly allocated Mach port in the remote task.
Completion of these steps results in the establishment of Mach ports, laying the groundwork for bidirectional communication.
In this section, the focus is on utilizing the execute primitive to establish basic memory read and write primitives. These initial steps are crucial for gaining more control over the remote process, though the primitives at this stage won't serve many purposes. Soon, they will be upgraded to more advanced versions.
The goal is to perform memory reading and writing using specific functions. For reading memory, functions resembling the following structure are used:
And for writing to memory, functions similar to this structure are used:
These functions correspond to the given assembly instructions:
A scan of common libraries revealed appropriate candidates for these operations:
Reading Memory: The property_getName() function from the Objective-C runtime library is identified as a suitable function for reading memory. The function is outlined below:
This function effectively acts like the read_func by returning the first field of objc_property_t.
Writing Memory: Finding a pre-built function for writing memory is more challenging. However, the _xpc_int64_set_value() function from libxpc is a suitable candidate with the following disassembly:
To perform a 64-bit write at a specific address, the remote call is structured as:
With these primitives established, the stage is set for creating shared memory, marking a significant progression in controlling the remote process.
The objective is to establish shared memory between local and remote tasks, simplifying data transfer and facilitating the calling of functions with multiple arguments. The approach involves leveraging libxpc and its OS_xpc_shmem object type, which is built upon Mach memory entries.
Memory Allocation:
Allocate the memory for sharing using mach_vm_allocate().
Use xpc_shmem_create() to create an OS_xpc_shmem object for the allocated memory region. This function will manage the creation of the Mach memory entry and store the Mach send right at offset 0x18 of the OS_xpc_shmem object.
Creating Shared Memory in Remote Process:
Allocate memory for the OS_xpc_shmem object in the remote process with a remote call to malloc().
Copy the contents of the local OS_xpc_shmem object to the remote process. However, this initial copy will have incorrect Mach memory entry names at offset 0x18.
Correcting the Mach Memory Entry:
Utilize the thread_set_special_port() method to insert a send right for the Mach memory entry into the remote task.
Correct the Mach memory entry field at offset 0x18 by overwriting it with the remote memory entry's name.
Finalizing Shared Memory Setup:
Validate the remote OS_xpc_shmem object.
Establish the shared memory mapping with a remote call to xpc_shmem_remote().
By following these steps, shared memory between the local and remote tasks will be efficiently set up, allowing for straightforward data transfers and the execution of functions requiring multiple arguments.
For memory allocation and shared memory object creation:
For creating and correcting the shared memory object in the remote process:
Remember to handle the details of Mach ports and memory entry names correctly to ensure that the shared memory setup functions properly.
Upon successfully establishing shared memory and gaining arbitrary execution capabilities, we have essentially gained full control over the target process. The key functionalities enabling this control are:
Arbitrary Memory Operations:
Perform arbitrary memory reads by invoking memcpy() to copy data from the shared region.
Execute arbitrary memory writes by using memcpy() to transfer data to the shared region.
Handling Function Calls with Multiple Arguments:
For functions requiring more than 8 arguments, arrange the additional arguments on the stack in compliance with the calling convention.
Mach Port Transfer:
Transfer Mach ports between tasks through Mach messages via previously established ports.
File Descriptor Transfer:
Transfer file descriptors between processes using fileports, a technique highlighted by Ian Beer in triple_fetch.
This comprehensive control is encapsulated within the threadexec library, providing a detailed implementation and a user-friendly API for interaction with the victim process.
Ensure proper use of memcpy() for memory read/write operations to maintain system stability and data integrity.
When transferring Mach ports or file descriptors, follow proper protocols and handle resources responsibly to prevent leaks or unintended access.
By adhering to these guidelines and utilizing the threadexec library, one can efficiently manage and interact with processes at a granular level, achieving full control over the target process.
Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE) Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)
