[CRIU] Ghost file: no such file or directory

[Tycho Andersen]

On Wed, Mar 30, 2016 at 06:15:20PM +0300, Pavel Emelyanov wrote:
> On 03/30/2016 06:10 PM, Tycho Andersen wrote:
> > On Wed, Mar 30, 2016 at 06:04:04PM +0300, Pavel Emelyanov wrote:
> >> On 03/30/2016 05:59 PM, Tycho Andersen wrote:
> >>> On Wed, Mar 30, 2016 at 05:50:32PM +0300, Pavel Emelyanov wrote:
> >>>> On 03/30/2016 05:46 PM, Tycho Andersen wrote:
> >>>>> On Wed, Mar 30, 2016 at 05:26:15PM +0300, Pavel Emelyanov wrote:
> >>>>>> On 03/30/2016 03:33 PM, Federico Reghenzani wrote:
> >>>>>>> Hi all!
> >>>>>>>
> >>>>>>> We have problem restoring Open MPI daemons with child processes that uses shared memory:
> >>>>>>>
> >>>>>>> (00.022447)    255: Opening ghost file 0x3 for tmp/openmpi-sessions-root at roaster-vm3_0/60995/1/shared_mem_pool.roaster-vm3.1
> >>>>>>> (00.022479)    255: Error (files-reg.c:139): Can't open ghost file //tmp/openmpi-sessions-root at roaster-vm3_0/60995/1/shared_mem_pool.roaster-vm3.1.cr.3.ghost: No such file or directory
> >>>>>>
> >>>>>> Can you check whether the \dirname of this path exists?
> >>>>>> I mean this -- //tmp/openmpi-sessions-root at roaster-vm3_0/60995/1/
> >>>>>>
> >>>>>> Presumably this is the case when not only the file was removed, but also
> >>>>>> some dir components. And we've fixed it only in 2.0.
> >>>>>
> >>>>> The users I had report it were using post 2.0, so there's some other
> >>>>> bug here. I'm trying to reproduce now again, but not having any luck
> >>>>> :(
> >>>>
> >>>> Do they also see the ENOENT errno from the open(O_CREAT) call?
> >>>
> >>> Which call do you mean here? I don't see anything close to this that
> >>> does an O_CREAT on dump.
> >>
> >> Yes, because the error is on restore :) For 1.8 this is files-reg.c create_ghost()'s
> > 
> > Oh, derp, I misread the error. I've seen failures reported on dump
> > here:
> > 
> > https://github.com/xemul/criu/blob/master/criu/files-reg.c#L628
> 
> Ouch! And what was the errno?! I can hardly imagine the reason for a regular
> file open failure via proc :(

ENOENT, yeah, I was confused as well, and haven't been able to
reproduce it either :(

Tycho