BUG: Python test #24 failure due to munmap()

[drakenclimber]

The python live test, 24-live-arg_allow.py, started failing on
Fedora 34 with kernel 5.13.9-200.fc34.x86_64. To fix this, allow
the munmap() syscall in the test's seccomp filter.

Signed-off-by: Tom Hromatka tom.hromatka@oracle.com

[drakenclimber]

@pcmoore - I ran into this test failure on step #14 of the release process.

I'm not sure yet if it's an issue on this particular machine, but I was able to reproduce it both in the git sandbox and the unpacked dist-gzip tarball.

Thoughts?

Oh, and I'll obviously fix it in main first, then cherry-pick the fix into the release-2.5 branch. Just wanted to get this out there.

[drakenclimber]

Without this change:

$ ./regression -m python -b 24-live-arg_allow -T live
=============== Tue Aug 31 04:10:59 PM MDT 2021 ===============
Regression Test Report ("regression -m python -b 24-live-arg_allow -T live")
 batch name: 24-live-arg_allow
 test mode:  python
 test type:  live
live
Test 24-live-arg_allow%%001-00001 result:   FAILURE 24-live-arg_allow 1 ALLOW rc=159
Regression Test Summary
 tests run: 1
 tests skipped: 0
 tests passed: 0
 tests failed: 1
 tests errored: 0
============================================================

With this change:

$ ./regression -m python -b 24-live-arg_allow -T live
=============== Tue Aug 31 04:11:31 PM MDT 2021 ===============
Regression Test Report ("regression -m python -b 24-live-arg_allow -T live")
 batch name: 24-live-arg_allow
 test mode:  python
 test type:  live
live
Test 24-live-arg_allow%%001-00001 result:   SUCCESS
Regression Test Summary
 tests run: 1
 tests skipped: 0
 tests passed: 1
 tests failed: 0
 tests errored: 0
============================================================

[coveralls]

Coverage remained the same at 89.487% when pulling 33cfbec on drakenclimber:issues/fix-test24 into 01a7e98 on seccomp:release-2.5.

[drakenclimber]

It looks like python 3.9 invokes munmap() on quit.

python 3.8.10 works on Ubuntu 20.10:

$ python --version
Python 3.8.10

$ ./regression -T live -m python -b 24-li
ve-arg_allow
=============== Wed Sep  1 15:25:03 UTC 2021 ===============
Regression Test Report ("regression -T live -m python -b 24-live-arg_allow")
 batch name: 24-live-arg_allow
 test mode:  python
 test type:  live
live
Test 24-live-arg_allow%%001-00001 result:   SUCCESS
Regression Test Summary
 tests run: 1
 tests skipped: 0
 tests passed: 1
 tests failed: 0
 tests errored: 0
============================================================

But Python 3.9.7 fails. No other changes made to the system:

$ python --version
Python 3.9.7

$ ./regression -T live -m python -b 24-live-arg_allow
=============== Wed Sep  1 15:25:57 UTC 2021 ===============
Regression Test Report ("regression -T live -m python -b 24-live-arg_allow")
 batch name: 24-live-arg_allow
 test mode:  python
 test type:  live
live
Test 24-live-arg_allow%%001-00001 result:   FAILURE 24-live-arg_allow 1 ALLOW rc=1
Regression Test Summary
 tests run: 1
 tests skipped: 0
 tests passed: 0
 tests failed: 1
 tests errored: 0
============================================================

[pcmoore]

Not sure why I do not see this in my nightly testing on Rawhide ... ? Regardless, it looks reasonable to me, thanks for catching this!

Acked-by: Paul Moore <paul@paul-moore.com>

[drakenclimber]

Not sure why I do not see this in my nightly testing on Rawhide ... ? Regardless, it looks reasonable to me, thanks for catching this!

That's really strange. As I said above, I was able to reproduce it on Fedora 34 with Python 3.9.6 and Ubuntu 20.10 with Python 3.9.7.

I would love to add Fedora (or something similar) to the Github Actions CI, but their supported distro list is rather scant. There is a way to add self-hosted runners (and we've had success in libcgroup using them), but it does put a small sysadmin burden on us.

[drakenclimber]

Thanks, @pcmoore

Merged into main in commit 4f34c6e
Merged into release-2.5 in commit 9b07575