Skip to contentSkip to conversation

selftests/input: add test to cover len > maxlen in bits_to_user()

PATCH In review Dana Elfassy 2 participants 2 messages Jun 21, 2023 – Jun 22, 2023 Kselftest

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 selftests/input: add test to cover len > maxlen in bits_to_user()

In order to cover this case, setting 'maxlen = 0', with the following explanation:
EVIOCGKEY is executed from evdev_do_ioctl(), which is called from evdev_ioctl_handler().
evdev_ioctl_handler() is called from 2 functions, where by code coverage, only the first one is in use.
‘compat’ is given the value ‘0’ [1].
Thus, the condition [2] is always false.
This means ‘len’ always equals a positive number [3]
‘maxlen’ in evdev_handle_get_val [4] is defined locally in evdev_do_ioctl() [5], and is sent in the variable 'size' [6]

[1] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1281
[2] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L705
[3] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L707
[4] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L886
[5] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1155
[6] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1141

---
Changes in v2:
- Added following note about the patch's dependency

This patch depends on '[v3] selftests/input: Introduce basic tests for evdev ioctls' [1] sent to the ML.
[1] https://patchwork.kernel.org/project/linux-input/patch/[email protected]/

 tools/testing/selftests/input/evioc-test.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
tools/testing/selftests/input/evioc-test.c +19 -0
index ad7b93fe39cf..b94de2ee5596 100644
@@ -234,4 +234,23 @@ TEST(eviocsrep_set_repeat_settings)
 	selftest_uinput_destroy(uidev);
 }
 
+TEST(eviocgkey_get_global_key_state)
+{
+	struct selftest_uinput *uidev;
+	int rep_values[2];
+	int rc;
+
+	memset(rep_values, 0, sizeof(rep_values));
+
+	rc = selftest_uinput_create_device(&uidev);
+	ASSERT_EQ(0, rc);
+	ASSERT_NE(NULL, uidev);
+
+	/* ioctl to create the scenario where len > maxlen in bits_to_user() */
+	rc = ioctl(uidev->evdev_fd, EVIOCGKEY(0), rep_values);
+	ASSERT_EQ(0, rc);
+
+	selftest_uinput_destroy(uidev);
+}
+
 TEST_HARNESS_MAIN
Signature · 1 line
2.41.0
#1 lore raw

Hi Dana,

Thank you for the patch.

Missatge de Dana Elfassy <[email protected]> del dia dc., 21 de juny 2023 a les 10:30:

Quoted 13 lines

In order to cover this case, setting 'maxlen = 0', with the following explanation:
EVIOCGKEY is executed from evdev_do_ioctl(), which is called from evdev_ioctl_handler().
evdev_ioctl_handler() is called from 2 functions, where by code coverage, only the first one is in use.
‘compat’ is given the value ‘0’ [1].
Thus, the condition [2] is always false.
This means ‘len’ always equals a positive number [3]
‘maxlen’ in evdev_handle_get_val [4] is defined locally in evdev_do_ioctl() [5], and is sent in the variable 'size' [6]

Like the comment in my previous patch I think this is understandable for someone that has some context, but I am not sure it is understood for people that don't have such context. So I'd try to rephrase and explain in a more plain way. I.e.

selftests/input: introduce a test for the EVIOCGKEY ioctl

This patch introduces a specific test case for the EVIOCGKEY ioctl. The test covers the case where len > maxlen in the EVIOCGKEY(sizeof(keystate)), keystate) ioctl.

Quoted 7 lines

[1] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1281
[2] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L705
[3] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L707
[4] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L886
[5] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1155
[6] https://elixir.bootlin.com/linux/v6.2/source/drivers/input/evdev.c#L1141

Also, usually is not a good idea to add links in commit messages reference to links that can easily end as dead links.

Quoted 16 lines

---
Changes in v2:
- Added following note about the patch's dependency

This patch depends on '[v3] selftests/input: Introduce basic tests for evdev ioctls' [1] sent to the ML.
[1] https://patchwork.kernel.org/project/linux-input/patch/[email protected]/

 tools/testing/selftests/input/evioc-test.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
tools/testing/selftests/input/evioc-test.c +0 -0
index ad7b93fe39cf..b94de2ee5596 100644

tools/testing/selftests/input/evioc-test.c TEST(eviocsrep_set_repeat_settings) · line 240

Quoted 7 lines
tools/testing/selftests/input/evioc-test.c +4 -0
@@ -234,4 +234,23 @@ TEST(eviocsrep_set_repeat_settings)
        selftest_uinput_destroy(uidev);
 }
 
+TEST(eviocgkey_get_global_key_state)
+{
+       struct selftest_uinput *uidev;
+       int rep_values[2];

nit: rep_values sounds like repeat values, I'd use a variable name more similar to what we really need to pass to the ioctl, like keystate. Also I think this can be simply

int keystate;

tools/testing/selftests/input/evioc-test.c TEST(eviocsrep_set_repeat_settings) · line 244

Quoted 4 lines
tools/testing/selftests/input/evioc-test.c +4 -0
+       int rc;
+
+       memset(rep_values, 0, sizeof(rep_values));
+

and then, this memset is not really needed.

tools/testing/selftests/input/evioc-test.c TEST(eviocsrep_set_repeat_settings) · line 245

Quoted 1 line
tools/testing/selftests/input/evioc-test.c +1 -0
+       rc = selftest_uinput_create_device(&uidev);

This one without extra arguments need to be:

        rc = selftest_uinput_create_device(&uidev, -1);

otherwise the vararg loop is going to keep the room warm for no reason.

tools/testing/selftests/input/evioc-test.c TEST(eviocsrep_set_repeat_settings) · line 252

Quoted 7 lines
tools/testing/selftests/input/evioc-test.c +7 -0
+       ASSERT_EQ(0, rc);
+       ASSERT_NE(NULL, uidev);
+
+       /* ioctl to create the scenario where len > maxlen in bits_to_user() */
+       rc = ioctl(uidev->evdev_fd, EVIOCGKEY(0), rep_values);
+       ASSERT_EQ(0, rc);
+

while we are here, and as it should be easy, maybe we can also extend this to cover the normal workflow to get the global key status?

Quoted 7 lines
tools/testing/selftests/input/evioc-test.c +3 -1
+       selftest_uinput_destroy(uidev);
+}
+
 TEST_HARNESS_MAIN
--

2.41.0

Thanks,
Enric