Fix race condition in legacy process stdio code

[kevpar]

HcsCreateProcess returns a set of stdio handles for the newly created
process. A while ago, we used to cache these handles and return them
the first time stdio handles were requested for the process, and then
get new handles via HcsGetProcessInfo for each subsequent request. At
some point, this code was cleaned up to instead always return the
original set of handles as non-closable (for new callers) and always get
new handles via HcsGetProcessInfo (for legacy callers, who required
closable handles).

However, this change introduced a race condition for legacy callers,
where in the case of a short lived container process, the container
could have terminated between when it was started and when the
orchestrator requested stdio handles. This led to ERROR_NOT_FOUND
being returned from HcsGetProcessInfo.

This change addresses this by returning the original handles the first
time stdio handles are requested, and then calling HcsGetProcessInfo for
every subsequent request (just as it used to work a while ago).

Signed-off-by: Kevin Parsons [email protected]

[jstarks]

Looks good with some suggestions.

[Iristyle]

I'm not sure if this PR is responsible for a new error I'm seeing or not -- but something in between nightly docker 2019-09-25 and nightly docker 2019-12-04 is causing errors on container exit under LCOW with a hcsshim::System::waitBackground message like this now:

PS C:\agent\_work\13\s> docker run -it hadolint/hadolint hadolint --version
Haskell Dockerfile Linter v1.17.3-1-g3fab409
time="2019-12-16T16:45:25-08:00" level=error msg="Error waiting for container: failed to shutdown container: container 33acef4c0b8c4edc15192cf81df85189a7b09d961ca893ac0437811c2625967c encountered an error during hcsshim::System::waitBackground: failure in a Windows system call: The virtual machine or container with the specified identifier is not running. (0xc0370110): subsequent terminate failed container 33acef4c0b8c4edc15192cf81df85189a7b09d961ca893ac0437811c2625967c encountered an error during hcsshim::System::waitBackground: failure in a Windows system call: The virtual machine or container with the specified identifier is not running. (0xc0370110)"

This is the last change to roll in that modified system.go so I figured I'd start here @kevpar