| Message ID | 20220722171534.3576419-4-dlatypov@google.com |
|---|---|
| State | Accepted |
| Commit | e562e309d1d4ac05457c1454b6007071f13b5684 |
| Headers | show |
| Series | [v2,1/5] kunit: string-stream: Simplify resource use | expand |
On Fri, Jul 22, 2022 at 1:15 PM Daniel Latypov <dlatypov@google.com> wrote: > > kunit_kfree() can only work on data ("resources") allocated by KUnit. > > Currently for code like this, > > void *ptr = kmalloc(4, GFP_KERNEL); > > kunit_kfree(test, ptr); > kunit_kfree() will segfault. > > It'll try and look up the kunit_resource associated with `ptr` and get a > NULL back, but it won't check for this. This means we also segfault if > you double-free. > > Change kunit_kfree() so it'll notice these invalid pointers and respond > by failing the test. > > Implementation: kunit_destroy_resource() does what kunit_kfree() does, > but is more generic and returns -ENOENT when it can't find the resource. > Sadly, unlike just letting it crash, this means we don't get a stack > trace. But kunit_kfree() is so infrequently used it shouldn't be hard to > track down the bad callsite anyways. > > After this change, the above code gives: > > # example_simple_test: EXPECTATION FAILED at lib/kunit/test.c:702 > > kunit_kfree: 00000000626ec200 already freed or not allocated by kunit > > Signed-off-by: Daniel Latypov <dlatypov@google.com> > Reviewed-by: David Gow <davidgow@google.com> Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
diff --git a/lib/kunit/test.c b/lib/kunit/test.c index 82019a78462e..c7ca87484968 100644 --- a/lib/kunit/test.c +++ b/lib/kunit/test.c @@ -698,18 +698,8 @@ static inline bool kunit_kfree_match(struct kunit *test, void kunit_kfree(struct kunit *test, const void *ptr) { - struct kunit_resource *res; - - res = kunit_find_resource(test, kunit_kfree_match, (void *)ptr); - - /* - * Removing the resource from the list of resources drops the - * reference count to 1; the final put will trigger the free. - */ - kunit_remove_resource(test, res); - - kunit_put_resource(res); - + if (kunit_destroy_resource(test, kunit_kfree_match, (void *)ptr)) + KUNIT_FAIL(test, "kunit_kfree: %px already freed or not allocated by kunit", ptr); } EXPORT_SYMBOL_GPL(kunit_kfree);
kunit_kfree() can only work on data ("resources") allocated by KUnit. Currently for code like this, > void *ptr = kmalloc(4, GFP_KERNEL); > kunit_kfree(test, ptr); kunit_kfree() will segfault. It'll try and look up the kunit_resource associated with `ptr` and get a NULL back, but it won't check for this. This means we also segfault if you double-free. Change kunit_kfree() so it'll notice these invalid pointers and respond by failing the test. Implementation: kunit_destroy_resource() does what kunit_kfree() does, but is more generic and returns -ENOENT when it can't find the resource. Sadly, unlike just letting it crash, this means we don't get a stack trace. But kunit_kfree() is so infrequently used it shouldn't be hard to track down the bad callsite anyways. After this change, the above code gives: > # example_simple_test: EXPECTATION FAILED at lib/kunit/test.c:702 > kunit_kfree: 00000000626ec200 already freed or not allocated by kunit Signed-off-by: Daniel Latypov <dlatypov@google.com> Reviewed-by: David Gow <davidgow@google.com> --- v1 -> v2: fix typo in commit desc. --- lib/kunit/test.c | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-)