[LTP] [PATCH V2] syscalls/openat2: New tests

[Cyril Hrubis]

Hi!
> > Wouldn't it be easier if we added these to the first test and keep this
> > one failures only?
> 
> I thought about that earlier and kept it this way as this file is only
> for testing the resolve flags. Else the success cases could be added
> to openat201.c and failure to openat203.c itself.
> 
> This also helps in understanding (or noticing), that the test only
> changes the value of the resolve flag and we get an error. The first
> test plays with a lot of variables and so it may not be best to do it
> there as it would be a bit less readable.

Okay then.

> > > +	/* Failure cases */
> > > +	{"resolve-no-xdev", "/proc/version", RESOLVE_NO_XDEV, EXDEV},
> > > +	{"resolve-no-magiclinks", "/proc/self/exe", RESOLVE_NO_MAGICLINKS, ELOOP},
> > > +	{"resolve-no-symlinks", FOO_SYMLINK, RESOLVE_NO_SYMLINKS, ELOOP},
> > > +	{"resolve-beneath", "/proc/version", RESOLVE_BENEATH, EXDEV},
> > > +	{"resolve-no-in-root", "/proc/version", RESOLVE_IN_ROOT, ENOENT},
> 
> > > +static struct tcase {
> > > +	const char *name;
> > > +	int dfd;
> > > +	const char *pathname;
> > > +	uint64_t flags;
> > > +	uint64_t mode;
> > > +	uint64_t resolve;
> > > +	struct open_how **how;
> > > +	size_t size;
> > > +	int exp_errno;
> > > +} tcases[] = {
> > > +	{"invalid-dfd", -1, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, 0, &how, sizeof(*how), EBADF},
> > > +	{"invalid-pathname", AT_FDCWD, NULL, O_RDONLY | O_CREAT, S_IRUSR, 0, &how, sizeof(*how), EFAULT},
> > > +	{"invalid-flags", AT_FDCWD, TEST_FILE, O_RDONLY, S_IWUSR, 0, &how, sizeof(*how), EINVAL},
> > > +	{"invalid-mode", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, -1, 0, &how, sizeof(*how), EINVAL},
> > > +	{"invalid-resolve", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, -1, &how, sizeof(*how), EINVAL},
> > > +	{"invalid-size-zero", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, 0, &how, 0, EINVAL},
> > > +	{"invalid-size-small", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, 0, &how, sizeof(*how) - 1, EINVAL},
> > > +	{"invalid-size-big", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, 0, &how, sizeof(*how) + 1, EFAULT},
> > > +	{"invalid-size-big-with-pad", AT_FDCWD, TEST_FILE, O_RDWR | O_CREAT, S_IRWXU, 0, (struct open_how **)&phow, sizeof(*how) + 1, E2BIG},
> > 
> > Here as well +8.
> 
> I kept this as 1 intentionally despite the fact that pad is 8 bytes
> long. The last 2 tests have size set to (sizeof(*how) + 1) now and the
> only difference is that we have provided pad of X number of bytes in
> one case and no pad in the other case. This gives us different error
> numbers based on difference in the pad available. If I use +8 here,
> then there are two factors which are different, the structure and the
> number of bytes we are sending in size and we can't be certain about
> why we got a different error number.

I do not get it, we can pass +8 to the EFAULT case as well, or whatever,
there is a whole page after the data.

What I would like to have here is that the size we pass to the kernel is
exactly the size of the data have have allocate, so that we catch
off-by-one accesses after the structure.



-- 
Cyril Hrubis
chrubis@suse.cz