Wasn't there an option to use iptables instead of the userland port proxy?

Enabling hairpin NAT via the daemon flag "--userland-proxy=false" prevents the daemon from spawning docker-proxy processes. This way the port mapping is achieved via just iptables (as Mrunal said).

Are there any reasons you can't disable userland-proxy? Could you try setting "--userland-proxy=false" and see if the daemon still has issues?

I believe that using the iptables rules prevents container to container communication.

could you tell me how you reproduce the blocked container to container communication?

Here is the information I was sent:

"""
Well, to explain it better, take a host that's on 192.168.0.1, running a Docker container listening on port 5000 (in the container) that's exposed on port 800 on the host.  Assume that the first container's Docker IP is 172.17.0.1.  Then, take another Docker container on the same host trying to connect to that exposed port;

APP=`docker run -d -p 800:5000/tcp training/webapp python app.py`
curl http://192.168.0.1:800/   # works, of course
docker run --rm -it image curl http://192.168.0.1:800/   # does not work
docker run --rm -it image curl http://172.17.0.1:5000/   # does work
docker stop $APP
docker rm $APP
"""

I haven't tried this myself yet however.

In docker-1.9.1-23.el7.x86_64, move to verified.

Just to be noted,
It's better not try `docker run -it --rm=true --name porttest -p 3000-4000 busybox` in  remote machine which used network is not strong enough.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHBA-2016-0536.html