KRETPROBES are broken since the commit 73f9b911faa7

[Adam Zabrocki]

Hello,

The recent KRETPROBE kernel changes introduced a potential NULL pointer
dereference bug when return handler is not set. The root of the problem
are the following commits related to "kprobes: Use rethook for kretprobe
if possible":
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f3a112c0c40dd96d53c8bdf3ea8d94d528f3b7b8
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73f9b911faa74ac5107879de05c9489c419f41bb


When user sets up a KRETPROBE, kernel eventually ends up here:

static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
                      struct pt_regs *regs)
{
    struct kretprobe *rp = (struct kretprobe *)data;
    struct kretprobe_instance *ri;
    struct kprobe_ctlblk *kcb;

    /* The data must NOT be null. This means rethook data structure is broken. */
    if (WARN_ON_ONCE(!data))
        return;

    __this_cpu_write(current_kprobe, &rp->kp);
    kcb = get_kprobe_ctlblk();
    kcb->kprobe_status = KPROBE_HIT_ACTIVE;

    ri = container_of(rh, struct kretprobe_instance, node);
    rp->handler(ri, regs);

    __this_cpu_write(current_kprobe, NULL);
}
NOKPROBE_SYMBOL(kretprobe_rethook_handler);

Unfortunately, rp->handler is not verified against NULL. If user does not
set up a return handler, KRETPROBE shouldn't call it (that's the defined
behavior in the documentation, quoting "Any or all handlers can be NULL").
However, kretprobe_rethook_handler function does not enforce such behavior.

I'm going to send a patch for it in a reply to this message.

Thanks,
Adam

-- 
pi3 (pi3ki31ny) - pi3 (at) itsec pl
http://pi3.com.pl