Re: [PATCH v3] lib: kunit: Provides a userspace memory context when tests are compiled as module

[Kees Cook <keescook@chromium.org>]

On Tue, Jul 21, 2020 at 02:40:36PM -0300, Vitor Massaru Iha wrote:
> KUnit test cases run on kthreads, and kthreads don't have an
> adddress space (current->mm is NULL), but processes have mm.
> 
> The purpose of this patch is to allow to borrow mm to KUnit kthread
> after userspace is brought up, because we know that there are processes
> running, at least the process that loaded the module to borrow mm.
> 
> This allows, for example, tests such as user_copy_kunit, which uses
> vm_mmap, which needs current->mm.

Ah! In the case of kunit starting before there IS a userspace...
interesting.

> Signed-off-by: Vitor Massaru Iha <vitor@massaru.org>
> ---
> v2:
>     * splitted patch in 3:
>         - Allows to install and load modules in root filesystem;
>         - Provides an userspace memory context when tests are compiled
>           as module;
>         - Convert test_user_copy to KUnit test;
>     * added documentation;
>     * added more explanation;
>     * added a missed test pointer;
>     * released mm with mmput();
> v3:
>     * rebased with last kunit branch
>     * Please apply this commit from kunit-fixes:
>         3f37d14b8a3152441f36b6bc74000996679f0998
> 
>  Documentation/dev-tools/kunit/usage.rst | 14 ++++++++++++++
>  include/kunit/test.h                    | 12 ++++++++++++
>  lib/kunit/try-catch.c                   | 15 ++++++++++++++-
>  3 files changed, 40 insertions(+), 1 deletion(-)
> ---
> diff --git a/Documentation/dev-tools/kunit/usage.rst b/Documentation/dev-tools/kunit/usage.rst
> index 3c3fe8b5fecc..9f909157be34 100644
> --- a/Documentation/dev-tools/kunit/usage.rst
> +++ b/Documentation/dev-tools/kunit/usage.rst
> @@ -448,6 +448,20 @@ We can now use it to test ``struct eeprom_buffer``:
> 
>  .. _kunit-on-non-uml:
> 
> +User-space context
> +------------------
> +
> +I case you need a user-space context, for now this is only possible through

typo: In case ...

> +tests compiled as a module. And it will be necessary to use a root filesystem
> +and uml_utilities.
> +
> +Example:
> +
> +.. code-block:: bash
> +
> +   ./tools/testing/kunit/kunit.py run --timeout=60 --uml_rootfs_dir=.uml_rootfs
> +
> +
>  KUnit on non-UML architectures
>  ==============================
> 
> diff --git a/include/kunit/test.h b/include/kunit/test.h
> index 59f3144f009a..ae3337139c65 100644
> --- a/include/kunit/test.h
> +++ b/include/kunit/test.h
> @@ -222,6 +222,18 @@ struct kunit {
>  	 * protect it with some type of lock.
>  	 */
>  	struct list_head resources; /* Protected by lock. */
> +	/*
> +	 * KUnit test cases run on kthreads, and kthreads don't have an
> +	 * adddress space (current->mm is NULL), but processes have mm.
> +	 *
> +	 * The purpose of this mm_struct is to allow to borrow mm to KUnit kthread
> +	 * after userspace is brought up, because we know that there are processes
> +	 * running, at least the process that loaded the module to borrow mm.
> +	 *
> +	 * This allows, for example, tests such as user_copy_kunit, which uses
> +	 * vm_mmap, which needs current->mm.
> +	 */
> +	struct mm_struct *mm;

I have a general concern that this will need more careful solving in the
future as there are likely to be many tests that need a userspace
context to operate sanely. But that's just a note; this solves the
specific case now.

>  };
> 
>  void kunit_init_test(struct kunit *test, const char *name, char *log);
> diff --git a/lib/kunit/try-catch.c b/lib/kunit/try-catch.c
> index 0dd434e40487..d03e2093985b 100644
> --- a/lib/kunit/try-catch.c
> +++ b/lib/kunit/try-catch.c
> @@ -11,7 +11,8 @@
>  #include <linux/completion.h>
>  #include <linux/kernel.h>
>  #include <linux/kthread.h>
> -
> +#include <linux/sched/mm.h>
> +#include <linux/sched/task.h>
>  #include "try-catch-impl.h"
> 
>  void __noreturn kunit_try_catch_throw(struct kunit_try_catch *try_catch)
> @@ -24,8 +25,17 @@ EXPORT_SYMBOL_GPL(kunit_try_catch_throw);
>  static int kunit_generic_run_threadfn_adapter(void *data)
>  {
>  	struct kunit_try_catch *try_catch = data;
> +	struct kunit *test = try_catch->test;
> +
> +	if (test != NULL && test->mm != NULL)
> +		kthread_use_mm(test->mm);
> 
>  	try_catch->try(try_catch->context);
> +	if (test != NULL && test->mm != NULL) {
> +		kthread_unuse_mm(test->mm);
> +		mmput(test->mm);
> +		test->mm = NULL;

This mmput() seems unbalanced... see below.

> +	}
> 
>  	complete_and_exit(try_catch->try_completion, 0);
>  }
> @@ -65,6 +75,9 @@ void kunit_try_catch_run(struct kunit_try_catch *try_catch, void *context)
>  	try_catch->context = context;
>  	try_catch->try_completion = &try_completion;
>  	try_catch->try_result = 0;
> +
> +	test->mm = get_task_mm(current);
> +
>  	task_struct = kthread_run(kunit_generic_run_threadfn_adapter,
>  				  try_catch,
>  				  "kunit_try_catch_thread");

Isn't there something that destroys a "struct kunit"? I would expect
that to perform the mmput(). Why is it up in the threadfn?

> 
> base-commit: d43c7fb05765152d4d4a39a8ef957c4ea14d8847
> --
> 2.26.2
> 

-- 
Kees Cook