libcontainer/specconv/spec_linux: Add support for (no)lazytime

[wking]

Part of catching runC up with the spec, which punts valid options to mount(8). Part of opencontainers/runtime-spec#771.

[wking]

Travis failure should be resolved by #1464. I'll rebase this PR on top of master if/when #1464 lands.

[wking]

Updated with 9707d1d → cdfb37e as requested by @cyphar. I'm still not sure if he wants me to pull in #1464 as well.

[wking]

Rebased onto master with cdfb37e → a57e41f to pick up #1464.

[wking]

The current Travis failure is a checkpoint issue. I saw a similar checkpoint issue on a different Go version in #1463, but I'm pretty sure this is unrelated to my PR because master recently failed with the same issue.

[hqhq]

Can you squash? It's a whole caching up for mount options and don't need to be a separate commit for each entry.

[wking]

On Thu, Jun 01, 2017 at 08:28:31PM -0700, Qiang Huang wrote: Can you squash?

Squashed with a57e41f → 4f81337. I'm really hoping that after all these consolidation requests nobody comes around and says “these all look good except for (no)acl” ;).

[cyphar]

Actually, let's drop noacl and rbind but keep the rest. 😉

In all seriousness, I'm worried by how many test failures we are having these days. @avagin do you think this is caused by the dropping of the wait_for_container code in our testing?

[crosbymichael]

@cyphar why rbind? that is extremely common

[crosbymichael]

Do not drop rbind

https://github.com/opencontainers/runc/blob/master/libcontainer/specconv/example.go#L216

[cyphar]

@crosbymichael It was a joke. 😉

[wking]

Triggering a Travis re-build now that #1484 has landed.

[wking]

Still get a Travis error, this time:

not ok 4 checkpoint --pre-dump and restore

apparently from trying to signal a finished process:

# runc list (status=0):
# ID             PID         STATUS      BUNDLE             CREATED                         OWNER
# test_busybox   0           stopped     /tmp/busyboxtest   2017-06-20T19:58:04.67322906Z   root
# runc kill test_busybox KILL (status=1):
# time="2017-06-20T19:58:05Z" level=error msg="container_linux.go:308: signaling init process caused \"os: process already finished\"\n" 
# container_linux.go:308: signaling init process caused "os: process already finished"

I think #1489 may be part of fixing that, possibly with a guard in the bats tests to avoid calling runc kill in the first place on stopped containers. It seems orthogonal to this PR in any case.

[avagin]

In all seriousness, I'm worried by how many test failures we are having these days. @avagin do you think this is caused by the dropping of the wait_for_container code in our testing?

If we remove a useless thing and start catching new bugs, it doesn't mean that the thing was not useless;).

[wking]

Triggering a Travis re-build now that #1489 has landed.

[wking]

And Travis is happy :).

[cyphar]

@wking By the way, we can retrigger travis build without closing and reopening PRs. 😉

[cyphar]

LGTM.

[crosbymichael]

LGTM

[AkihiroSuda]

This implementation might not be correct:

• mount: acl implementation is probably wrong #3738