Add tests for miscellaneous administrative system calls

[aniketp]

Summary

This revision introduces atf-c(3) tests for many unrelated administrative system calls.
Note: Most the syscalls are too critical to be tested in success mode.

List of Syscalls:

• ntp_adjtime(2)
• auditctl(2)
• acct(2)
• auditon(2)
• clock_settime(2)

// Only in failure mode

• reboot(2)
• quotactl(2)
• mount(2)
• nmount(2)
• swapon(2)
• swapoff(2)

Test Plan

Execute make && make install from test/sys/audit.
Execute kyua test from /usr/tests/sys/audit. All testcases should succeed.

[asomers]

tests/sys/audit/administrative.c

171 ↗

(On Diff #44061)

timebuff is used uninitialized.

267 ↗

(On Diff #44061)

What happens to the system's audit trail after this command? Does it keep recording the stuff in /var/audit?

656 ↗

(On Diff #44061)

Can't you test quotactl successfully using the non-destructive Q_GETQUOTASIZE, Q_GETQUOTA, or Q_SYNC commands?

685 ↗

(On Diff #44061)

Actually, you can mount a filesystem. nullfs would work, for example. But be sure to unmount it during the cleanup!

[aniketp]

tests/sys/audit/administrative.c

171 ↗

(On Diff #44061)

Is it not supposed to be filled by ntp_adjtime(2). That is what man-page suggested.

267 ↗

(On Diff #44061)

After I call auditctl(2) myself, auditing starts in my configured path rather than /var/audit.
But when the audit daemon is stopped and started. Process resumes in /var/audit

Actually:
When the auditd(8) is not already running, everything works fine after running the tests. But on the other side, auditing in /var/audit stops.

So basically, to make everything work normally, we'll have to stop and start auditd8) again. (I verified this behaviour using an independent program).

656 ↗

(On Diff #44061)

int size;
if (quotactl("/", Q_GETQUOTASIZE, 0, &size) < 0)
        perror("quotactl");

outputs:
quotactl: Invalid argument

From the man-page, it looks like the error is because the filesystem quota is not already enabled on my system

MAN
The cmd argument or the command type is invalid.  In
Q_GETQUOTASIZE, Q_GETQUOTA, Q_SETQUOTA, and Q_SETUSE,
quotas are not currently enabled for this file system.

[asomers]

tests/sys/audit/administrative.c

171 ↗

(On Diff #44061)

No, you're misreading it. That argument is provided from userland to the kernel.

267 ↗

(On Diff #44061)

Tough call. Do we skip the test if auditd is already running? Or do we run it anyway, but make sure to restart auditd in the cleanup? I'm leaning toward the latter, even though it will cause a short disruption to the system-wide audit trail. My reasoning is that none of the ATF tests should be run in a production environment, so a brief audit-trail disruption should be ok.

628 ↗

(On Diff #44061)

If we're going to temporarily disrupt the audit trail, then we may as well disrupt process accounting, too. You can check whether accounting is enabled with the kern.acct_configured sysctl. Then, during cleanup, either do "service accounting onestop" (if it wasn't enabled) or "service accounting restart" (if it was).

656 ↗

(On Diff #44061)

Does that even happen for Q_SYNC?

[aniketp]

• Add test for acct_success. Use sysctlbyname(3) to retrieve the accounting status
• Minor changes in auditctl_success test case cleanup

[asomers]

tests/sys/audit/administrative.c

346 ↗

(On Diff #44203)

Why go to so much effort not to enable accounting? It would be simpler to make acctpath a constant. It will be deleted anyway at the end of the test.

357 ↗

(On Diff #44203)

You can't do cleanup in the test body, because it won't run if the test fails. You need to move this to the cleanup function.

[aniketp]

Update the test case for acct_success

[asomers]

tests/sys/audit/administrative.c

335 ↗

(On Diff #44237)

It's not useful to document that the file is conditional; the condition one line above makes that clear. If you're going to document it, it should be to highlight its role in the cleanup function.

356 ↗

(On Diff #44237)

Pro tip: if you initialize filedesc2 to -1, then you can unconditionally close it and ignore the return value. However, in this case it would be better to move the close statement up to line 336, immediately after opening it.

[aniketp]

Shift the location of close(filedesc2) to right after opening it

[asomers]

tests/sys/audit/administrative.c

685 ↗

(On Diff #44061)

Are you still working on mounting a filesystem?

[aniketp]

tests/sys/audit/administrative.c

685 ↗

(On Diff #44061)

I've been trying out numerous variations of commands to see if they execute successfully for mount(2) and nmount(2).

Whenever I get the "type" right, i.e either "nullfs" or "tmpfs", Any other set of command outputs Operation not supported.
For the wrong types, I expectedly get invalid argument

[aniketp]

• Add tests for auditon(2), swapon(2) and swapoff(2)

Note: swapon(2) and swapoff(2) require block device, hence can't be tested
in success case since block devices are gone (from FreeBSD)

[aniketp]

• Add tests for clock_settime(2)

[asomers]

If you intend to add successful test cases for mount, swapon, etc later, then you shouldn't indicate them as TODOs in the comments. Remove the text about how mount can't be tested in success mode.

[aniketp]

• Remove placeholder comment about not testing mount(2), nmount(2) in success case

These will be dealt later on

[asomers]

tests/sys/audit/administrative.c

385 ↗

(On Diff #44531)

evclass is used uninitialized here.

[aniketp]

Initialize evclass to get the event-class mapping for auditon(2)

[asomers]

tests/sys/audit/administrative.c

386 ↗

(On Diff #44541)

You should initialize ec_class too, even though current kernel code doesn't use it. Otherwise, tools like Coverity will detect this as a use of uninitialized data. Anything will work. Either bzero, or an explicit value.

[aniketp]

• Initialize ec_class to 0