incorrect bind() syscall behavior if address already in use and container was resumed

Steps to Reproduce:

1. install latest CentOS 6.x
2. install latest OpenVZ for CentOS 6.x
3, create container, 32-bit or 64-bit, enter to container (for example, 123)
4. install latest stable nginx inside this container (with ip 172.22.22.123)
5. configure nginx "listen 172.22.22.123:80 default_server backlog=1024"
6. start nginx, enable nginx service autostart (chkconfig nginx on)
7. restart entire server (command "restart" in hardware node ssh)
8. after restart - login into container with nginx and run "strace nginx -t"

Actual Results:

EADDRINUSE only in listen() after successful bind() from "nginx -t" instance

# netstat -tunlp | grep :80
tcp 0 0 172.22.22.123:80 0.0.0.0:* LISTEN 596/nginx
tcp 0 0 127.0.0.1:80 0.0.0.0:* LISTEN 589/httpd

# strace nginx -t
bind(12, {sa_family=AF_INET, sin_port=htons(80), sin_addr=inet_addr("172.22.22.123")}, 16) = 0
listen(12, 1024) = -1 EADDRINUSE (Address already in use)

Expected Results:

EADDRINUSE in bind(), because address 172.22.22.123 already in use by 596/nginx

# netstat -tunlp | grep :80
tcp 0 0 172.22.22.123:80 0.0.0.0:* LISTEN 596/nginx
tcp 0 0 127.0.0.1:80 0.0.0.0:* LISTEN 589/httpd

# strace nginx -t
bind(12, {sa_family=AF_INET, sin_port=htons(80), sin_addr=inet_addr("172.22.22.123")}, 16) = -1 EADDRINUSE (Address already in use)

Additional Information:

temporary workaround exists: after restart container ("vzctl restart 123") bind() inside this container work as expected. but, only before next full server reboot. while rebooting - vz service suspend/resume all containers and this bind() bug already appears.

vzctl.log:
...
 vzctl : CT 123 : Checkpointing completed successfully
...
vzctl : CT 123 : Dump file /vz/dump/Dump.123 exists, trying to restore from it
vzctl : CT 123 : Restoring container ...
vzctl : CT 123 : Container is mounted
vzctl : CT 123 : Adding IP address(es): 172.22.22.123
vzctl : CT 123 : Setting CPU units: 1000
vzctl : CT 123 : Container start in progress...
vzctl : CT 123 : Restoring completed successfully
...
============ now BUG can be observed ===================
...
vzctl : CT 123 : Restarting container
vzctl : CT 123 : Stopping container ...
vzctl : CT 123 : Container was stopped
vzctl : CT 123 : Container is unmounted
vzctl : CT 123 : Starting container...
vzctl : CT 123 : Container is mounted
vzctl : CT 123 : Adding IP address(es): 172.22.22.123
vzctl : CT 123 : Setting CPU units: 1000
vzctl : CT 123 : Container start in progress...
...
============== no more BUG, bind() work as expected =============

this bug first appears with recent 2.6.32-042stabXXXX kernels, after vzctl change behaviour to checkpoint/restore containers during server restart.

with old vzctl (without VE_STOP_MODE=suspend)
bind() syscall BUG was not observed (suspend not used at all).