[LTP] fsconfig04: Check FSCONFIG_SET_PATH

[Andrea Cervesato]

Hi Wei,

> > > +	TEST(fsconfig(fd, FSCONFIG_SET_PATH, "journal_path", dev2, 0));
> > > +	if (TST_RET == -1) {
> > > +		if (TST_ERR == EOPNOTSUPP)
> > > +			tst_brk(TCONF, "fsconfig(FSCONFIG_SET_PATH) not supported");
> > > +		else
> > > +			tst_brk(TFAIL | TTERRNO, "fsconfig(FSCONFIG_SET_PATH) failed");
> > > +	}
> > 
> > Could this test be gated with .min_kver = "7.2"?
> > 
> > Before Linux 7.2, ext4 declared journal_path as a string parameter, so
> > FSCONFIG_SET_PATH is rejected with EINVAL rather than EOPNOTSUPP. Since
> > fsopen_supported_by_kernel() only gates fsopen at Linux 5.2, those kernels
> > reach the else branch and report TFAIL instead of TCONF.
> If fsconfig parameter is correct, return EINVAL should be an kernel bug.
> Current logic is correct.
> I guess AI mix wtih some old code with new API. But how to fix AI's rule is a
> question.

yeah, agent is wrong.

> > 
> > > +	SAFE_SSCANF(dev2, "/dev/%s", loop_name);
> > > +
> > > +	snprintf(path, sizeof(path), "/sys/block/%s/dev", loop_name);
> > > +	SAFE_FILE_SCANF(path, "%u:%u", &major, &minor);
> > 
> > Could the device number be obtained from stat(dev2).st_rdev instead?

what about this?

> > 
> > tst_find_free_loopdev() also supports /dev/loop/N and /dev/block/loopN.
> > Those paths produce loop/N or block/loopN here, causing the test to read a
> > nonexistent sysfs path and terminate with TBROK.
> > 
> > > +	SAFE_MKFS(dev0, tst_device->fs_type, mkfs_opts_set_journal_dev1, NULL);
> > > +}
> > > +
> > > +static void run(void)
> > > +{
> > > +	...
> > > +	TEST(fsconfig(fd, FSCONFIG_SET_PATH, "journal_path", dev2, 0));
> > 
> > Could each iteration reset dev0 to dev1 or alternate the source and target
> > journal devices?
> > 
> > The first iteration persists dev2 in dev0's superblock. Every subsequent
> > -i iteration requests dev2 again, so it no longer exercises a dynamic
> > journal-device change.

and this?

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
andrea.cervesato@suse.com