[PATCH 2/2] landlock: Selftests for truncate(2) support.
Mickaël Salaün
mic at digikod.net
Fri Jul 8 11:17:46 UTC 2022
Please use "selftests/landlock:" as subject prefix and without a final dot.
On 07/07/2022 22:06, Günther Noack wrote:
> These tests exercise the following scenarios:
>
> * File with Read, Write, Truncate rights.
Should we use a capital for access right names or does it come from Go? ;)
> * File with Read, Write rights.
> * File with Truncate rights.
> * File with no rights.
> * Directory with Truncate rights.
>
> For each of the scenarios, both truncate() and the open() +
> ftruncate() syscalls get exercised and their results checked.
>
> In particular, the test demonstrates that opening a file for writing
> is not enough to call truncate().
Looks good! According to my previous comment, O_TRUNC should be tested
if it is checked by the kernel.
>
> Signed-off-by: Günther Noack <gnoack3000 at gmail.com>
> ---
> tools/testing/selftests/landlock/fs_test.c | 80 ++++++++++++++++++++++
> 1 file changed, 80 insertions(+)
>
> diff --git a/tools/testing/selftests/landlock/fs_test.c b/tools/testing/selftests/landlock/fs_test.c
> index cb77eaa01c91..c3e48fd12b2b 100644
> --- a/tools/testing/selftests/landlock/fs_test.c
> +++ b/tools/testing/selftests/landlock/fs_test.c
> @@ -2237,6 +2237,86 @@ TEST_F_FORK(layout1, reparent_rename)
> ASSERT_EQ(EXDEV, errno);
> }
>
> +TEST_F_FORK(layout1, truncate)
Please move this test after the proc_pipe one.
> +{
> + const struct rule rules[] = {
You can add a first layer of rules to check truncate and ftruncate with
a ruleset not handling LANDLOCK_ACCESS_FS_TRUNCATE.
> + {
> + .path = file1_s1d1,
> + .access = LANDLOCK_ACCESS_FS_READ_FILE |
> + LANDLOCK_ACCESS_FS_WRITE_FILE |
> + LANDLOCK_ACCESS_FS_TRUNCATE,
> + },
> + {
> + .path = file2_s1d2,
> + .access = LANDLOCK_ACCESS_FS_READ_FILE |
> + LANDLOCK_ACCESS_FS_WRITE_FILE,
> + },
> + {
> + .path = file1_s1d2,
> + .access = LANDLOCK_ACCESS_FS_TRUNCATE,
> + },
Move this entry before file2_s1d2 to keep the paths sorted and make this
easier to read. You can change the access rights per path to also keep
their ordering though.
> + {
> + .path = dir_s2d3,
> + .access = LANDLOCK_ACCESS_FS_TRUNCATE,
> + },
> + // Implicitly: No access rights for file2_s1d1.
Comment to move after the use of file1_s1d1.
> + {},
> + };
> + const int ruleset_fd = create_ruleset(_metadata, ACCESS_ALL, rules);
Don't use ACCESS_ALL because it will change over time and we want tests
to be deterministic. You can use rules[0].access instead.
> + int reg_fd;
> +
> + ASSERT_LE(0, ruleset_fd);
> + enforce_ruleset(_metadata, ruleset_fd);
> + ASSERT_EQ(0, close(ruleset_fd));
> +
> + /* Read, write and truncate permissions => truncate and ftruncate work. */
It would be nice to have consistent comments such as: "Checks read,
write and truncate access rights: truncate and ftruncate work."
> + reg_fd = open(file1_s1d1, O_RDWR | O_CLOEXEC);
> + ASSERT_LE(0, reg_fd);
> + EXPECT_EQ(0, ftruncate(reg_fd, 10));
You should not use EXPECT but ASSERT here. I use EXPECT when an error
could block a test or when it could stop a cleanup (i.e. teardown).
> + EXPECT_EQ(0, ftruncate64(reg_fd, 20));
> + ASSERT_EQ(0, close(reg_fd));
> +
> + EXPECT_EQ(0, truncate(file1_s1d1, 10));
> + EXPECT_EQ(0, truncate64(file1_s1d1, 20));
> +
> + /* Just read and write permissions => no truncate variant works. */
> + reg_fd = open(file2_s1d2, O_RDWR | O_CLOEXEC);
> + ASSERT_LE(0, reg_fd);
> + EXPECT_EQ(-1, ftruncate(reg_fd, 10));
> + EXPECT_EQ(EACCES, errno);
> + EXPECT_EQ(-1, ftruncate64(reg_fd, 20));
> + EXPECT_EQ(EACCES, errno);
> + ASSERT_EQ(0, close(reg_fd));
> +
> + EXPECT_EQ(-1, truncate(file2_s1d2, 10));
> + EXPECT_EQ(EACCES, errno);
> + EXPECT_EQ(-1, truncate64(file2_s1d2, 20));
> + EXPECT_EQ(EACCES, errno);
> +
> + /* Just truncate permissions => truncate(64) works, but can't open file. */
> + ASSERT_EQ(-1, open(file1_s1d2, O_RDWR | O_CLOEXEC));
> + ASSERT_EQ(EACCES, errno);
> +
> + EXPECT_EQ(0, truncate(file1_s1d2, 10));
> + EXPECT_EQ(0, truncate64(file1_s1d2, 20));
> +
> + /* Just truncate permission on directory => truncate(64) works, but can't open file. */
> + ASSERT_EQ(-1, open(file1_s2d3, O_RDWR | O_CLOEXEC));
> + ASSERT_EQ(EACCES, errno);
> +
> + EXPECT_EQ(0, truncate(file1_s2d3, 10));
> + EXPECT_EQ(0, truncate64(file1_s2d3, 20));
> +
> + /* No permissions => Neither truncate nor ftruncate work. */
> + ASSERT_EQ(-1, open(file2_s1d1, O_RDWR | O_CLOEXEC));
> + ASSERT_EQ(EACCES, errno);
> +
> + EXPECT_EQ(-1, truncate(file2_s1d1, 10));
> + EXPECT_EQ(EACCES, errno);
> + EXPECT_EQ(-1, truncate64(file2_s1d1, 20));
> + EXPECT_EQ(EACCES, errno);
These tests are good!
> +}
> +
> static void
> reparent_exdev_layers_enforce1(struct __test_metadata *const _metadata)
> {
More information about the Linux-security-module-archive
mailing list