BUG: update the s390 and s390x munge functions

[drakenclimber]

The s390 and s390x munge functions were missing a few syscalls and that was causing failures in the s390x TravisCI run.

This is another step to resolving pull request #215

[coveralls]

Coverage decreased (-0.04%) to 92.701% when pulling 967c984 on drakenclimber:pull/215 into 5abc3b1 on seccomp:master.

[pcmoore]

Do we also need to add the following?

• semop
• shmat
• shmdt
• shmget
• shmctl

[drakenclimber]

Good question. I'll run them through the s390x VM to see if they're needed. If so, I'll likely add them to the test to ensure they are properly verified in the future.

[drakenclimber]

Do we also need to add the following?

* semop
* shmat
* shmdt
* shmget
* shmctl

Based upon Arnd's commit that was a catalyst for much of this work, I would think they should be munged as well.

But I can't create a real-world test scenario where the munge of the above shm*() syscalls is required. (At least on my s390x machine, that is. Note that I don't have console access and grubby isn't accepting my attempts to change to a newer kernel, so I'm limited to a really old kernel. I've tried hacking /usr/include/asm/unistd.h by hand to influence the behavior, but it hasn't affected any outcomes.)

Strangely, I can get test 37 to fail if the msg*() syscalls aren't munged, but the munging (or not munging) of shm*() doesn't affect the test.

I'm torn. I think you are correct that they should also be munged, but I can't prove it. Thoughts, @pcmoore?

[giuseppe]

please ignore my review, I've clicked by mistake while I was still checking if it is something I broke (and my knowledge is quite limited to make any real review)

[pcmoore]

Doing the munge to a pseudo syscall should be harmless as the libseccomp code will Do The Right Thing and generate the necessary rules to handle both cases, where not doing the munge will cause only the new direct-wired syscall rule to be created.

That said, we should also take a closer look at s390/s390x to make sure that the *_syscall_demux() and *_syscall_mux() are also correct.

[pcmoore]

please ignore my review, I've clicked by mistake while I was still checking if it is something I broke (and my knowledge is quite limited to make any real review)

:)

Thanks for looking, but this was broken long before your changes ;)

[drakenclimber]

The various shm*() functions were already in both the s390x and s390 munge/demunge code, so they should be good to go.

semop() is not supported by s390 or s390x.

I did add semtimedop() support to the s390 mux/demux functions as it had not yet been enabled.

[pcmoore]

Looks good to me, merged at HEAD 0e3db6a. Thanks!