Bug#500645: linux-image-2.6.26-1-openvz-686: OpenVZ checkpointing does not work

[Faidon Liambotis <paravoid@debian.org>]

Package: linux-image-2.6.26-1-openvz-686
Version: 2.6.26-6
Severity: normal

2.6.26-6 supposedly added OpenVZ checkpointing support -- and indeed
the kernel option was enabled and /proc/cpt exists.

I was unable, however, to perform an online migration or even a
checkpoint/restore cycle on both 686 and amd64 kernels on different
machines:

	hn:~# vzctl chkpnt 201
	  Setting up checkpoint...
	  	suspend...
	  	dump...
	  	kill...
	  VE is unmounted
	  Checkpointing completed succesfully
	hn:~# vzctl restore 201
	  Restoring VE ...
	  Starting VE ...
	  VE is mounted
	  	undump...
	  Adding IP address(es): <removed>
	  Setting CPU units: 1000
	  Configure meminfo: 49152
	  Error: undump failed: Address already in use
	  Restoring failed:
	  Error: open_listening_socket: bind: -98
	  Error: rst_sockets: open_listening_socket: -98
	  Error: rst_sockets: -98
	  VE start failed
	  Stopping VE ...
	  VE was stopped
	  VE is unmounted
	hn:~# dmesg |tail -5
	  [ 1013.441803] CT: 201: started
	  [ 1014.711018] CPT ERR: cdeae800,201 :open_listening_socket: bind: -98
	  [ 1014.712378] CPT ERR: cdeae800,201 :rst_sockets: open_listening_socket: -98
	  [ 1014.713732] CPT ERR: cdeae800,201 :rst_sockets: -98
	  [ 1014.725277] CT: 201: stopped

The problem seems to be with bind(), obviously.
I tried checkpointing a container that had all network services shutdown and
it worked. I also tried veth instead of venet with no luck.

I've searched the Internet and upstream's bugzilla with no relevant
results.

I can provide you with access on a kvm guest that has this problem in case
you want to debug it further. Note that I reproduced this on real
machines as well, so this has nothing to do with kvm.

Regards,
Faidon