[LTP] [PATCH V5 05/10] syscalls/fsconfig: New tests

[Petr Vorel <pvorel@suse.cz>]

> Add tests to check working of fsconfig() syscall.

...
> +++ b/testcases/kernel/syscalls/fsconfig/fsconfig01.c
...
> +static void run(void)
> +{
> +	int fd, fsmfd;
> +
> +	TEST(fd = fsopen(tst_device->fs_type, 0));
> +	if (fd == -1)
> +		tst_brk(TBROK | TERRNO, "fsopen() failed");
> +
> +	TEST(fsconfig(fd, FSCONFIG_SET_FLAG, "rw", NULL, 0));
> +	if (TST_RET == -1) {
> +		tst_res(TFAIL | TERRNO, "fsconfig() failed");
> +		goto out;
Because there is .test_all (single run), although which is called N times for
each filesystem (.all_filesystems = 1), but cleanup is called for each of them
it's I'll move SAFE_CLOSE(fd) to cleanup function and use here (and on the rest
of calls) return (trying to avoid goto when possible).
I'll change it before merge.

> +
> +	TEST(fsmfd = fsmount(fd, 0, 0));
> +	if (fsmfd == -1) {
> +		tst_res(TBROK | TERRNO, "fsmount() failed");
> +		goto out;
BTW This needs to be now tst_res(TFAIL) (this patchset now does not compile).

> +	TEST(fsconfig(fd, FSCONFIG_SET_FD, "sync", NULL, 0));
> +	if (TST_RET == -1) {
> +		if (TST_ERR == EOPNOTSUPP) {
> +			tst_res(TCONF, "fsconfig(): FSCONFIG_SET_FD not supported");
> +		} else {
> +			tst_res(TFAIL | TERRNO, "fsconfig() failed");
> +			goto out;
> +		}
> +	}

I get TCONF for all fsconfig01 results, while I'm using 5.5.5-1-default:

tst_test.c:1290: INFO: Testing on ext2
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with ext2 opts='' extra opts=''
mke2fs 1.45.5 (07-Jan-2020)
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:43: CONF: fsconfig(): FSCONFIG_SET_PATH not supported
fsconfig01.c:53: CONF: fsconfig(): FSCONFIG_SET_PATH_EMPTY not supported
fsconfig01.c:63: CONF: fsconfig(): FSCONFIG_SET_FD not supported
fsconfig01.c:92: PASS: fsconfig() passed
tst_test.c:1290: INFO: Testing on ext3
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with ext3 opts='' extra opts=''
mke2fs 1.45.5 (07-Jan-2020)
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:43: CONF: fsconfig(): FSCONFIG_SET_PATH not supported
fsconfig01.c:53: CONF: fsconfig(): FSCONFIG_SET_PATH_EMPTY not supported
fsconfig01.c:63: CONF: fsconfig(): FSCONFIG_SET_FD not supported
fsconfig01.c:92: PASS: fsconfig() passed
tst_test.c:1290: INFO: Testing on ext4
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with ext4 opts='' extra opts=''
mke2fs 1.45.5 (07-Jan-2020)
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:43: CONF: fsconfig(): FSCONFIG_SET_PATH not supported
fsconfig01.c:53: CONF: fsconfig(): FSCONFIG_SET_PATH_EMPTY not supported
fsconfig01.c:63: CONF: fsconfig(): FSCONFIG_SET_FD not supported
fsconfig01.c:92: PASS: fsconfig() passed
tst_test.c:1290: INFO: Testing on xfs
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with xfs opts='' extra opts=''
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:92: PASS: fsconfig() passed
tst_test.c:1290: INFO: Testing on btrfs
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with btrfs opts='' extra opts=''
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:43: CONF: fsconfig(): FSCONFIG_SET_PATH not supported
fsconfig01.c:53: CONF: fsconfig(): FSCONFIG_SET_PATH_EMPTY not supported
fsconfig01.c:63: CONF: fsconfig(): FSCONFIG_SET_FD not supported
fsconfig01.c:92: PASS: fsconfig() passed
tst_test.c:1290: INFO: Testing on vfat
tst_mkfs.c:89: INFO: Formatting /dev/loop0 with vfat opts='' extra opts=''
tst_test.c:1227: INFO: Timeout per run is 0h 05m 00s
fsconfig01.c:43: CONF: fsconfig(): FSCONFIG_SET_PATH not supported
fsconfig01.c:53: CONF: fsconfig(): FSCONFIG_SET_PATH_EMPTY not supported
fsconfig01.c:63: CONF: fsconfig(): FSCONFIG_SET_FD not supported
fsconfig01.c:92: PASS: fsconfig() passed

Not yet merged man page [1] (I reposted David Howells commit) there is
explanation for EOPNOTSUPP: The command given by cmd was not valid.

First, I suspected "sync" option is wrong. But looking at kernel sources it's
really not implemented:

fs/fsopen.c
	if (fc->ops == &legacy_fs_context_ops) {
		switch (cmd) {
		case FSCONFIG_SET_BINARY:
		case FSCONFIG_SET_PATH:
		case FSCONFIG_SET_PATH_EMPTY:
		case FSCONFIG_SET_FD:
			ret = -EOPNOTSUPP;
			goto out_f;
		}
	}

fs/fs_context.c
	/* TODO: Make all filesystems support this unconditionally */
	init_fs_context = fc->fs_type->init_fs_context;
	if (!init_fs_context)
		init_fs_context = legacy_init_fs_context;
...
/*
 * Initialise a legacy context for a filesystem that doesn't support
 * fs_context.
 */
static int legacy_init_fs_context(struct fs_context *fc)
{
	fc->fs_private = kzalloc(sizeof(struct legacy_fs_context), GFP_KERNEL);
	if (!fc->fs_private)
		return -ENOMEM;
	fc->ops = &legacy_fs_context_ops;
	return 0;
}

Code coming from v5.1-rc1 f3a09c92018a91ad0981146a4ac59414f814d801 introduce
fs_context methods [2]. Other patchsets here [3] shows there is some support for
fs_context in VFS. So I wonder how to achieve not end up with legacy context.

> +++ b/testcases/kernel/syscalls/fsconfig/fsconfig02.c
> +  {"set-path-key", &fd, FSCONFIG_SET_PATH, NULL, "/dev/sda1", &aux_fdcwd, EINVAL},
...
> +  {"set-path-aux", &fd, FSCONFIG_SET_PATH, "sync", "/dev/sda1", &aux_minus1, EINVAL},
/dev/sda1 is valid on some hosts, but invalid on others. Shouldn't we use
/dev/foo instead?

...
> +	temp_fd = open("testfile", O_RDWR | O_CREAT, 01444);
> +	if (temp_fd == -1)
> +		tst_res(TBROK, "Can't obtain temp_fd, open() failed");
Here needs to be now tst_brk(TBROK).
Again I'll change it before merge.

...

Kind regards,
Petr

[1] https://marc.info/?l=linux-man&m=158109737907972&w=2
[2] https://patchwork.kernel.org/patch/10820207/
[3] https://patchwork.kernel.org/project/linux-security-module/list/?series=82377