BUG: some tests fail on RHEL-6

[pcmoore]

We need to attempt a test run on a RHEL-6 system.

[The-Mule]

Test are running, there is just a minor issue in the test driver - auditctl is expected in /usr/sbin but located in /sbin, I will patch it. Most of the test failed, I have to investigate that. We should also introduce some logic to distinguish between rhel-6, rhel-7 and fedora tests.

[The-Mule]

Well I did my homework and check all the tests in RHEL6. Details are as follows.

Working

• syscalls_file
• user_msg

Broken but Compatible

• exec_execve - on RHEL6, it might take some time to populate events, test sometimes fails and sometimes passes, I hate adding sleeps to tests, but in this case adding 1s sleep before events before ausearch call makes it green

Broken

• file_* - are these tests for directory watches? Is it supposed to be working on RHEL6? Rules are added correctly but no syscalls are audited at all

Generally Broken

• login_tty - this test is broken for all distributions at the moment.

[pcmoore]

Great, thanks for looking into this, I think we can move this from Q to BUG.

[rgbriggs]

Ok, I tracked down the problem with the file_* tests. The tempdir and tempfile functions don't honour the TEMPLATE assignment, so the intended file never gets created or tested. Testing with $filename = "$dir/$key";system("touch $filename"); works fine.

[rgbriggs]

The login_tty test doesn't work on RHEL6 because that feature/bugfix hasn't been backported to RHEL6.

[rgbriggs]

Not directly related to this issue, but in the process of debugging the tests, I found it difficult to parse the log output. It would be nice if there was a clear delimiter between each subdir test so when failures happen it is easier to see what part of the output goes with which test case.

[The-Mule]

Indeed, I usually call this when debugging: # HARNESS_VERBOSE=1 HARNESS_OPTIONS="c" SUBDIRS="file_create" make -e test First option makes standard output visible, second just colorize summary/results from the rest. When parsing records within while loop, it is always better to print each line - during the test execution, it is not visible anyway and with HARNESS_VERBOSE=1 it is really helpful.

…

On Fri, Dec 2, 2016 at 12:38 PM, Richard Guy Briggs < ***@***.***> wrote: Not directly related to this issue, but in the process of debugging the tests, I found it difficult to parse the log output. It would be nice if there was a clear delimiter between each subdir test so when failures happen it is easier to see what part of the output goes with which test case. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#19 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKfnBJ9v1xYS5liOdVP2zsjjh2eNv4lOks5rEAM8gaJpZM4HdUtX> .

[rgbriggs]

Here's a quick and dirty tempfile TEMPLATE workaround for RHEL6: rgbriggs@4d0fddc

[pcmoore]

Here's a quick and dirty tempfile TEMPLATE workaround for RHEL6: rgbriggs@4d0fddc

Thanks guys for the help on this.

One of the things that is nice about tempfile() is the automatic cleanup when UNLINK is true. @rgbriggs you mention that RHEL-6.x doesn't honor the tempfile() TEMPLATE parameter, does it still generate a file if we omit the TEMPLATE parameter? The name/location of the temp file generally shouldn't be important to us (assuming it is created in a valid temp/writable location, but I think that is a safe assumption).

[rgbriggs]

It still generates a file while seemingly ignoring the TEMPLATE parameter, so I assume it would generate the same pattern of file with the TEMPLATE parameter missing. The file was generated in a safe place. Knowing that the TEMPLATE parameter was ignored (or incorrect syntax of the supplied parameter?) then I would cast some doubt that the UNLINK parameter was honoured either.

[pcmoore]

It should be easy enough to verify if the UNLINK worked.

[rgbriggs]

The UNLINK parameter was also ignored. There are temp files hanging around in /tmp/.

[pcmoore]

Thanks for checking, but that's a bummer. We definitely need to find an alternate method of creating, and cleaning up, temporary files that works on RHEL-6.x forward (yes, I'm intentionally ignoring RHEL-5.x).

[rgbriggs]

The quick and dirty patch from #19 (comment) does clean up after itself. The tempdir command does reap the files from the touch system command.

[pcmoore]

So the TEMPLATE parameter works for tempdir() but not tempfile()? I didn't think that was the case based on your earlier comment, perhaps I misunderstood.

I don't see any rm/unlink actions in the quick n' dirty patch so the only way the temp files could be deleted is if $dir was deleted. What am I missing?

[rgbriggs]

The TEMPLATE parameter appears to not work for either tempdir() or tempfile(), but it appears unlink works or defaults to on for tempdir() but not for tempfile(), which surprised me.

[pcmoore]

Huh, yeah, I wouldn't have guess that either. What a mess.

This unpredictability makes me wonder about the difficulty of doing our own temp file management.

[rgbriggs]

Well, for other tests I've just been making tempfiles based on the unique key generated for the test and cleaning up after myself.

[pcmoore]

That may be the only reasonable solution, my only concern is cleanup in the case of abnormal termination, e.g. kill the test via Ctrl-C. That said, I'm not even sure the existing tempdir()/tempfile() solution handles this correctly, and I'm also not sure how much we should ultimately care about this; as long as everything is created under /tmp, artifacts under corner cases may be acceptable.

[The-Mule]

It actually might be good to have some "clean-up stack" which is called when test exits or is killed / aborted. For instance, when you create a temporary file, you just add a removing call-back onto clean-up stack. Callbacks from the stack are then called by some exit subroutine or from a signal handler. Does it make sense? I can open an issue and write something for demonstration and we can then decide whether it is useful or not...?

[pcmoore]

@The-Mule yes, that makes sense to me (and I think what the original tempfile/tempdir code did), I just don't currently know how to do signal/exit handlers in perl; if you can write up a demonstration I think that would be very helpful.

[pcmoore]

It's not clear how much effort we want to put into RHEL-6.x at this point so let's close this out for now.