xdk kernel module

August 28, 2025 ยท View on GitHub

The xdk kernel module helps simulating vulnerabilities in the kernel, tracking function calls and testing exploitation primitives.

Usage

  1. Include the xdk_device.h header file, so the structures used by the API will be available in the program.

  2. Open the /dev/xdk device for read and write:

  int fd = open("/dev/xdk", O_RDWR);

  1. Send commands to the module via ioctl:

    int ret;
xdk_message msg = { .length = 1024 };
if ((ret = ioctl(fd, ALLOC_BUFFER, &msg)) != SUCCESS) {
    printf("xdk_dev command failed with %u\n", ret);
}

    The return value of the ioctl is one of the enum xdk_errors values:

    • zero is SUCCESS
    • non-zero values are errors

Examples

Examples how to use the module can be found in the test/xdk_dev_test.c file or the below in the command-specific documentation.

Supported commands

The supported commands are listed in enum xdk_cmd. Currently the following commands are supported:

ALLOC_BUFFER: Kernel buffer allocation

Allocates a kernel buffer and optionally copies data from the user-space into the newly allocated buffer.

Usage

int ret;

xdk_message msg = { .length = 1024, .gfp_account = 1 };
msg.data = malloc(msg.length);
memset(msg.data, 0x41, msg.length);

if ((ret = ioctl(fd, ALLOC_BUFFER, &msg)) == SUCCESS) {
    printf("Kernel address of the allocated buffer: %p\n", msg.kernel_ptr);
}

If .gfp_account is set to 1, then it allocates the memory with the GFP_KERNEL_ACCOUNT flag, otherwise it allocates with GPF_KERNEL.

KFREE: Kernel memory deallocation

Calls kfree() on the argument.

Usage

uint64_t kernel_address = 0xffff...;
if (ioctl(fd, KFREE, kernel_address) == SUCCESS)
    printf("Kernel address (0x%lx) was kfree()'d.\n", kernel_address);

KASLR_LEAK: Get kASLR base address

Returns the kASLR adjusted virtual address of the _text symbol (kASLR base address for .text and some other sections).

Usage

uint64_t kaslr_base;
if (ioctl(fd, KASLR_LEAK, &kaslr_base) == SUCCESS)
    printf("kaslr base: %lx\n", kaslr_base);

SYM_ADDR: Get any kallsyms symbol address

Returns the kASLR adjusted virtual address of the requested symbol.

Note: without CONFIG_KALLSYMS_ALL, only function addresses are available.

Usage

sym_addr sym_addr = { .symbol_name = "core_pattern" };
if (ioctl(fd, SYM_ADDR, &sym_addr) == SUCCESS)
    printf("core_pattern address: %lx\n", sym_addr.symbol_address);

WIN_TARGET: Get a RIP target address

Gets the address of a kernel function which can be called from a ROP chain or via RIP control primitive and it prints the following text into the kernel log:

[    1.535654] ...
[    1.538542] xdk_dev: win_target was called.
[    1.538542] 
[    1.538542] !!! YOU WON !!! 
[    1.538542] 
[    1.541079] ...

Usage

uint64_t win_target;
if (ioctl(fd, WIN_TARGET, &kaslr_base) == SUCCESS)
    printf("win_target: %lx\n", win_target);

PRINTK:

Logs the user-supplied text into the kernel logs (it is useful for maintaining the order of message coming from both user and kernel-space).

Usage

ioctl(fd, PRINTK, "hello world");

Which is shown in the logs as:

[    1.513672] hello world

ARB_READ:

Copies memory from arbitrary kernel address to user-space.

Usage

sym_addr sym_addr = { .symbol_name = "core_pattern" };
if (ioctl(fd, SYM_ADDR, &sym_addr) != SUCCESS) return;

uint8_t buffer[32];
xdk_message msg = { .length = sizeof(buffer), .data = buffer, .kernel_addr = sym_addr.symbol_addr };
if (ioctl(fd, ARB_READ, &msg) == SUCCESS) {
    printf("current core_pattern = '%s', should be same as:\n", buffer);
    system("cat /proc/sys/kernel/core_pattern");
}

ARB_WRITE:

Copies memory from user-space to arbitrary kernel address.

Usage

sym_addr sym_addr = { .symbol_name = "core_pattern" };
if (ioctl(fd, SYM_ADDR, &sym_addr) != SUCCESS) return;

char new_core_pattern[] = "|/tmp/run_as_root";
xdk_message msg = { .length = sizeof(new_core_pattern), .data = new_core_pattern, .kernel_addr = sym_addr.symbol_addr };
if (ioctl(fd, ARB_WRITE, &msg) == SUCCESS) {
    printf("core_pattern was successfully overwritten, the new value is:\n");
    system("cat /proc/sys/kernel/core_pattern");
}

RIP_CONTROL: Simulate a RIP control primitive

Simulates a RIP control primitive by setting the generic CPU registers to user-provided values and then executes one of the following assembly instructions (defined in enum rip_action):

  • JMP_RIP: jmp r15
    • R15 is set to rip_control_args.rip
  • CALL_RIP: call r15
    • R15 is set to rip_control_args.rip
  • RET: ret
    • RSP is expected to be set to the address of a fake stack containing e.g. a ROP chain

The following registers can be set: RAX, RBX, RCX, RDX, RSI, RDI, RBP, RSP, R8, R9, R10, R11, R12, R13, R14. (R15 can only be set for the RET action, otherwise it's value is ignored and stores the call / jmp target instead.)

It needs to be explicitly specified which registers to set via the rip_control_args.regs_to_set field which is a bitwise OR of the enum regs_to_set values (see examples below).

Usage: call a valid RIP target

uint64_t win_target;
if (ioctl(fd, WIN_TARGET, &win_target) != SUCCESS) return;

rip_control_args rip = { .action = CALL_RIP, .rip = win_target };
if (ioctl(fd, RIP_CONTROL, &rip) == SUCCESS)
    printf("User-space program execution continues after executing our target function in the kernel...\n");

Which results in the following logs:

[    1.398862] xdk_dev: rip_control: action=0x2, rsp=0x0, value@rsp=0x0, regs_to_set=0x0, rip=0xffffffffc02e8218
[    1.401394] xdk_dev: win_target was called.
[    1.401394] 
[    1.401394] !!! YOU WON !!! 
[    1.401394] 
[    1.403947] xdk_dev: rip_control, after asm
User-space program execution continues after executing our target function in the kernel...

Usage: set RDI and jump to address

rip_control_args rip = { .action = JMP_RIP, .regs_to_set = RDI, .rip = 0xffffff4141414141, .rdi = 0x4242424242424242 };
ioctl(fd, RIP_CONTROL, &rip);

Which results in the following crash (RIP and RDI set to the provided values):

[    1.434875] RIP: 0010:0xffffff4141414141                                       // RIP == 0xffffff4141414141
[    1.435939] Code: Unable to access opcode bytes at 0xffffff4141414117.
[    1.437659] RSP: 0018:ffffa6f680423ae8 EFLAGS: 00010246
[    1.439027] RAX: 0000000000000001 RBX: ffffa6f680423b28 RCX: 0000000000000000
[    1.440895] RDX: 0000000000000000 RSI: 0000000000000002 RDI: 4242424242424242  // RDI == 0x4242424242424242

Usage: jump to a ROP chain by setting RSP

// create ROP chain
uint64_t rop[128];
rop[0] = 0xffffff4141414141;

// store ROP chain in kernel memory
xdk_message msg = { .length = sizeof(rop), .data = &rop };
if (ioctl(fd, ALLOC_BUFFER, &msg) != SUCCESS) return;

// jump to ROP chain by setting RSP to the ROP chain's kernel memory address
rip_control_args rip = { .action = RET, .regs_to_set = RSP, .rsp = msg.kernel_addr };
ioctl(fd, RIP_CONTROL, &rip);

Which results in a crash like (ROP chain item is executed):

[    1.381932] xdk_dev: rip_control: action=0x3, rsp=0xffff9e5d025e7c00, value@rsp=0xffffff4141414141, regs_to_set=0x80, rip=0x0
[    1.384801] BUG: unable to handle page fault for address: ffffff4141414141
...
[    1.396223] RIP: 0010:0xffffff4141414141
[    1.397276] Code: Unable to access opcode bytes at 0xffffff4141414117.
[    1.398991] RSP: 0018:ffff9e5d025e7c08 EFLAGS: 00010246

Compilation

Automatically with image_runner

If you run ./run.sh with --custom-modules=xdk_device, the module will be compiled and loaded automatically.

Manually for image_runner targets

Run ./compile_custom_modules.sh (kernelctf|ubuntu) <release-name> xdk_device and after the compilation process, the compiled module can be found at rootfs/custom_modules/xdk_device.ko.

Manually for custom kernels

Compile your kernel normally and then execute the following command from the same directory where you compiled your kernel (replace <image_runner_dir> with the root directory of the image_runner):

make M=<image_runner_dir>/../third_party/kernel_modules/xdk_device modules