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

[Viresh Kumar]

On 06-03-20, 15:46, Cyril Hrubis wrote:
> > diff --git a/testcases/kernel/syscalls/openat2/openat202.c b/testcases/kernel/syscalls/openat2/openat202.c
> > new file mode 100644
> > index 000000000000..504878277f7e
> > --- /dev/null
> > +++ b/testcases/kernel/syscalls/openat2/openat202.c
> > @@ -0,0 +1,86 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Copyright (c) 2020 Viresh Kumar <viresh.kumar@linaro.org>
> > + *
> > + * openat2() tests with various resolve flags.
> > + */
> > +#include "tst_test.h"
> > +#include "lapi/openat2.h"
> > +
> > +#define FOO_SYMLINK "foo_symlink"
> > +
> > +static struct open_how *how;
> > +
> > +static struct tcase {
> > +	const char *name;
> > +	const char *pathname;
> > +	uint64_t resolve;
> > +	int exp_errno;
> > +} tcases[] = {
> > +	/* Success cases */
> > +	{"open /proc/version", "/proc/version", 0, 0},
> > +	{"open magiclinks", "/proc/self/exe", 0, 0},
> > +	{"open symlinks", FOO_SYMLINK, 0, 0},
> 
> 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.

> > +	/* 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.

-- 
viresh