[CRIU] Checkpoint fails intermittently

[Pavel Emelyanov]

On 10/06/2014 02:12 AM, Frederico Araujo wrote:
> Dear list,
> 
> I've notice that sometimes checkpoint fails with the following error (please see attached log):
> 
> /Error (sk-tcp.c:66): Can't turn TCP repair mode ON: Operation not permitted/
> 
> However, I'm executing criu as sudo. Could it be that the file descriptor for the socket being
> dumped is set as RO or immutable? In which cases this error would occur? 

It also happens if the socket we're turning repair on is not in closed/established
state. Since this error happens not every time, I think I know why this happens.

The TCP socket dump goes in 3 steps

1. collect

  Here we talk to kernel's tcp_diag subsys fetching the info about tcp sockets.
  At this stage we get the state the socket is in and next two steps only occur
  if the state is closed/established.

2. lock

  Here we block the packets with netfilter

3. dump

  Here we turn repair on and get the socket state into image file


So, if between steps 1 and 2 the FIN packet arrives the socket would get turned
into one of the closing states and we will fail to turn repair ON on it.  

Thanks,
Pavel