This repo implements a kexec()-style system call for the PS4 Orbis kernel (FreeBSD derivative). This is designed to boot a Linux kernel directly from FreeBSD.
This is not an exploit. It is useless without some mechanism of injecting code into the PS4 OS kernel.
To build a kexec.bin relocatable binary using the supplied Makefile, just type make. This will also build a kexec.a archive. You can either use the binary directly, or link the archive into your own project.
If you link kexec.a with your own code, you need to supply the two symbols _start and _end in your linker script, as kernel_init() will try to remap all pages covered by that range as RWX (to make global variable accesses work). Alternatively, you can add -DDO_NOT_REMAP_RWX to CFLAGS to disable this feature, if you have already taken care of page permissions for the code.
If you use a compiler toolchain that have a special prefix you can declare it by passing TOOLCHAIN_PREFIX option to the Makefile like this:
The code is designed to be completely standalone. There is a single entry point:
int kexec_init(void *early_printf, sys_kexec_t *sys_kexec_ptr);
Simply call kexec_init(NULL, NULL). This will locate all the required kernel symbols and install the sys_kexec system call. The syscall is registered as number 153 by default (you can change this in kexec.h). The return value is 0 on success, or negative on error.
You may pass something other than NULL as early_printf. In that case, that function will be used for debug output during early symbol resolution, before printf is available.
Since PS4 3.55(?), KASLR(Kernel Address Space Layout Randomization) is enabled by default, symtab also disappears in newer kernel, we have to hardcode offsets for some symbols. Currently we use the early_printf given by user to caculate the base address of kernel, then relocate all the symbols from the kernel base. You could enable this feature like this:
make CFLAG='-DPS4_4_00 -DKASLR -DNO_SYMTAB'
If you do not want to call the syscall from userspace, you can pass the address of a function pointer as sys_kexec_ptr. kexec_init will write to it the address of sys_kexec, so you can invoke it manually (see kexec.h for its prototype and how the arguments are passed).
If you are using the standalone kexec.bin blob, then the kexec_init function is always located at offset 0, so simply call the base address of the blob. Don't forget to pass two NULL arguments (or the appropriate pointers).
The injected sys_kexec system call takes (userspace) pointers to the kernel and initramfs blobs, their sizes, and a pointer to the (null-terminated) command line string. From userspace, this looks like this:
int kexec(void *kernel_image, size_t image_size,
void *initramfs, size_t initramfs_size,
const char *cmdline);
// syscall() usage:
syscall(153, kernel_image, image_size, initramfs, initramfs_size, cmdline);
kexec() will load the kernel and initramfs into memory, but will not directly boot them. To boot the loaded kernel, shut down the system. This can be accomplished by pressing the power button, but can also be done more quickly and reliably from userspace with the following sequence of system calls (this kills userspace quickly but still does a controlled filesystem unmount):
int evf = syscall(540, "SceSysCoreReboot");
syscall(546, evf, 0x4000, 0);
// should be syscall(37, 1, 30) but only tested via kill symbol
Note that this software should be loaded into kernel memory space. If you are running kernel code from userland mappings, you should either switch to kernel mappings or separately copy kexec.bin to a location in kernel address space. While syscalls or exploit code may run properly from userland, the shutdown hook will not, as it will be called from a different process context.
kernel_init() will automatically find the Orbis OS kernel and resolve all necessary symbols to work. There are no static symbol dependencies. If DO_NOT_REMAP_RWX is not defined (the default), it will also patch pmap_protect to disable the W^X restriction.
In addition to loading the user-supplied initramfs, kexec will locate the Radeon firmware blobs inside Orbis OS, extract them, convert them to a format suitable for Linux, and append them as an additional initramfs cpio image to the existing initramfs. This avoids the need to distribute the Radeon firmware blobs. The radeon module, when compiled into the kernel, will automatically load this firmware on boot. Note however that most typical initramfs scripts will wipe the initramfs contents while pivoting to the real system, so if you compile radeon as a module you may not be able to access the firmware after boot. To cover that case, add some code to your initramfs /init script to copy the firmware to a tmpfs mounted on the real filesystem:
# assuming real root FS is mounted on /mnt
mkdir -p /mnt/lib/firmware/radeon
mount -t tmpfs none /mnt/lib/firmware/radeon
cp /lib/firmware/radeon/* /mnt/lib/firmware/radeon/
# now switch_root to /mnt
This avoids having to permanently store copies of the Radeon firmware, which isn't really necessary for most use cases.
There is significant debug logging available, which will appear on the system UART. Most of the code relies on the kernel printf implementation, and therefore you should patch out the UART output blanker to see it. The final code that runs on the boot CPU before booting the kernel uses direct UART writes and is not affected by the blanking feature of Orbis OS.