Details

Reviewers

asomers
rwatson
gnn

Commits

rS335679: audit(4): add tests for the process-control audit class

Summary

This revision introduces a new (and final) test-program for syscalls categorized within
process-control audit class. A lot of these syscalls don't have tests for either
success or failure case due to various reasons

List of System calls

Both cases

rfork(2)
chdir(2)
fchdir(2)
chroot(2)
getresuid(2)
getresgid(2)
setpriority(2)
setgroups(2)
setpgrp(2)
setrlimit(2)
setlogin(2)
mlock(2)
munlock(2)
minherit(2)
rtprio(2)
profil(2)
ktrace(2)
ptrace(2)

Success case

fork(2)
umask(2)
setuid(2)
setgid(2)
seteuid(2)
setegid(2)

Test Plan

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

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

aniketp created this revision.Jun 22 2018, 3:45 PM

Harbormaster completed remote builds in B17543: Diff 44306.Jun 22 2018, 3:45 PM

aniketp edited the summary of this revision. (Show Details)Jun 22 2018, 3:45 PM

aniketp set the repository for this revision to rS FreeBSD src repository - subversion.

Herald added a subscriber: imp. · View Herald TranscriptJun 22 2018, 3:45 PM

Add test for successful invocation of setlogin(2)

Harbormaster completed remote builds in B17545: Diff 44309.Jun 22 2018, 4:04 PM

aniketp edited the summary of this revision. (Show Details)Jun 22 2018, 4:05 PM

asomers requested changes to this revision.Jun 22 2018, 5:39 PM

asomers added inline comments.

tests/sys/audit/process-control.c
72 ↗	(On Diff #44309)	You could move check_audit up a line. In fact, it isn't really necessary to call wait at all.
75 ↗	(On Diff #44309)	FYI, calling exit(3) after a fork is surprisingly risky. That's because it tries to do various cleanup activities. Usually those cleanup activities will work. But if the parent process was multithreaded, then the child will be single-threaded, and its only thread will be a copy of the parent thread that forked. That means that mutexes could end up leaked and other nastiness. That's why it's a common pattern to instead call _exit(2) after forking. It doesn't matter here, because your program is single-threaded and doesn't have any mutexes or other stuff that could interfere with the child. Just keep it in mind for the future.
158 ↗	(On Diff #44309)	Kyua is smart enough to chdir back to the test directory for cleanup. So you could simplify this test by simply chdir()ing to "/".
660 ↗	(On Diff #44309)	It's fine to set priority to a hardcoded value like 0. This process will exit pretty soon anyway.
707 ↗	(On Diff #44309)	gidset is used uninitialized.
740 ↗	(On Diff #44309)	Don't forget to add this testcase.
1156 ↗	(On Diff #44309)	Probably safer to use pid instead of hardcoding 0.

This revision now requires changes to proceed.Jun 22 2018, 5:39 PM

Changes according to the inline comments

Harbormaster completed remote builds in B17557: Diff 44334.Jun 23 2018, 2:26 AM

asomers added inline comments.Jun 23 2018, 6:48 PM

tests/sys/audit/process-control.c
699 ↗	(On Diff #44334)	Safer to pass the full gidset to setgroups instead of only the first group.

Update ngroups for setgroups_success as 5 instead of 100

Harbormaster completed remote builds in B17572: Diff 44355.Jun 23 2018, 7:26 PM

aniketp added inline comments.Jun 23 2018, 7:27 PM

tests/sys/audit/process-control.c

699 ↗

(On Diff #44334)

@asomers setgroups is peculiar in this regard.
If I pass in entire gidset array as the maximum limit it will never let the audit record complete itself.

The piece of code here:

	case AUE_SETGROUPS:
		if (ARG_IS_VALID(kar, ARG_GROUPSET)) {
			for(ctr = 0; ctr < ar->ar_arg_groups.gidset_size; ctr++)
			{
				tok = au_to_arg32(1, "setgroups",
				    ar->ar_arg_groups.gidset[ctr]);
				kau_write(rec, tok);
			}
		}
		break;

Loops till all gidset elements are inserted in the audit record. So for 100 such groups, I get an incomplete audit record and tests fail.
This doesn't happen with 1 or 5 as the limit.

With ngroups = 100;

Looking for 'setgroups.*1844.*ret.*success' in 'header,1868,11,setgroups(2),0,Sat Jun 23 19:15:30 2018, + 663 msecargument,1,0x0,setgroupsargument,1,0x0,setgroupsargument,1,0x5,setgroupsargument,1,0x0,setgroupsargument,1,0xe7913e01,setgroupsargument,1,0x8078a49c,setgroupsargument,1,0x635090,setgroupsargument,1,0x8,setgroupsargument,1,0xffffd480,setgroupsargument,1,0x7fff,setgroupsargument,1,0x77bf89,setgroupsargument,1,0x8,setgroupsargument,1,0x635090,setgroupsargument,1,0x8,setgroupsargument,1,0xe7913ef0,setgroupsargument,1,0x8078a49c,setgroupsargument,1,0xe7913ef0,setgroupsargument,1,0x8078a49c,setgroupsargument,1,0x1,setgroupsargument,1,0x0,setgroupsargument,1,0x635250,setgroupsargument,1,0x8,setgroupsargument,1,0xe7913ef0,setgroupsargument,1,0x8078a49c,setgroupsargument,1,0x1,setgroupsargument,1,0x0,setgroupsargument,1,0x64d010,setgroupsargument,1,0x8,setgroupsargument,1,0x635090,setgroupsargument,1,0x8,setgroupsargument,1,0xffffd3e0,setgroupsargument,1,0x7fff,setgroupsargument,1,0xffffd3a0,setgroupsargument,1,0x7fff,setgroupsargument,1,0x7d6869,setgro����'

process-control:setgroups_success  ->  failed: setgroups.*1844.*ret.*success not found in auditpipe within the time limit

Note: for the sake of the test case, I'm using ngroups as 5 instead of 100.

asomers accepted this revision.Jun 23 2018, 7:57 PM

asomers added inline comments.

tests/sys/audit/process-control.c
699 ↗	(On Diff #44334)	Looks like you're running up against the 1024 character limit imposed in utils.c:52. So, not a bug in audit(4). A 5-group limit is fine. BTW, you might see a 16-group limit some places. That comes from NFSv2 and NFSv3, which set that limit in the RPCs.