mocking init_task ?

[Kees Cook <keescook@chromium.org>]

Hi,

I want to make a unit test for kill_something_info(), as there is a patch
to fix a bug with it not working as expected under a specific process
tree arrangement[1]. This seems like a great candidate for a unit test:
given a specific state, return a specific result. Emboldened, I applied
the "kunit: Support redirecting function calls" series[2], preparing to
mock group_send_sig_info(), and ran head-long into for_each_process()
... which uses the address of the global init_task:

#define for_each_process(p) \
        for (p = &init_task ; (p = next_task(p)) != &init_task ; )

:(

I'm curious what you think might be the right approach to mock
init_task, or for_each_process(), so I can apply unit tests to some of
the "simple" process tree walkers...

One idea I had was using the "kunit: Provide a static key to check if
KUnit is actively running tests" series[3], and do something like this:

#ifndef CONFIG_KUNIT
#define init_task_ptr	&init_task
#else
#define init_task_ptr	({					\
		struct task_struct *task = &init_task;		\
		if (static_branch_unlikely(&kunit_running)) {	\
			struct kunit *test;			\
			test = current->kunit_test;		\
			if (test->mock_init_task)		\
				task = test->mock_init_task;	\
		}						\
		task;						\
	})
#endif

#define for_each_process(p) \
        for (p = init_task_ptr ; (p = next_task(p)) != init_task_ptr ; )

And then tests can hang a mock init_task off the test? It seems really
horrible, but there is a LOT of global state in the kernel, so I figured
I had to start somewhere? :P

Thoughts?

-Kees

[1] https://lore.kernel.org/lkml/20221122161240.137570-1-pskocik@gmail.com/
[2] https://lore.kernel.org/lkml/20220910212804.670622-1-davidgow@google.com/
[3] https://lore.kernel.org/lkml/20221125084306.1063074-1-davidgow@google.com/

-- 
Kees Cook