Bug 1390922

+++ This bug was initially created as a clone of Bug #1390921 +++

Description of problem:
When using docker by docker in a systems CI configuration there is no guaranteed consistent GID that is safe to use for the socket to be mounted into the container.

The RHEL package just does a groupadd -r dockerroot and the Fedora package doesn't even create such a group in the scripts.

It's useful, and important in a docker by docker situation, to provide the group to the socket (via the -G option in DOCKER_OPTS) to lock down socket usage without requiring a sudo configuration as well (for example in the Jenkins docker pipeline world it looks for docker specifically and can't use an alias of docker=sudo docker).

However since there is not an allocated GID it's possible, even likely, for the container to end up with a GID different from the host and then using -v to mount results in a clash and permission issues - this is even more likely when building a container on one host that might be used on another so the Dockerfile cannot have a lookup of the GID during build for a RUN groupadd.

Version-Release number of selected component (if applicable):
docker-1.10.3-46.el7.centos.14.x86_64

How reproducible:
always

Steps to Reproduce:
1. dnf install docker
2. getent group | grep dockerroot
3. 

Actual results:
no group on fedora, a dynamically allocated group on rhel/centos

Expected results:
consistent group gid