Re: [Xen-devel] xm save workaround

[John Croft <john@xxxxxxxxxxxxxxxx>]

Calling xc_interface_open returns a file descriptor which has had it's 
close-on-exec flag set.
Disassembling the code in xc_interface_open in libxenctrl.so as shipped 
in the FC5 release has two calls to fcntl just after the open.. GET_FD,  
the flags, or in 1 and then SET_FD . This isn't in the xc_misc.c in the 
repository, so I don't know how it's getting there.
John

Ewan Mellor wrote:

> On Sat, Mar 25, 2006 at 09:04:40AM +0000, John Croft wrote:
>
>  
>
> > Hi all,
> >
> > This is something I posted earlier on Xen-Users. I understand it is a known 
> > problem so you may be interested in this workaround.
> > cheers,
> > John
> >
> >
> > Hi John-
> >
> > Thanks a lot, that did the trick.  Have you sent this to the Xen-Devel
> > list?  If not, I bet they would be interested.
> >
> > Anyway, thanks again!
> >
> > -Patrick
> >
> > On 3/24/06, John Croft <john@xxxxxxxxxxxxxxxx> wrote:
> >
> >    
> >
> > > > Hi Patrick,
> > > >
> > > > I'm wondering if this is related to the problem I had and posted on the
> > > > xen-users list with a work-around. The symptoms are similar in that
> > > > xm_save fails but the debug output is slightly different, but that may
> > > > be due to environmental differences.
> > > >
> > > > original post follows.
> > > > John
> > > >
> > > > Hi all,
> > > >
> > > > Just installed Xen from the base FC5 released on Monday.
> > > >
> > > > It all seems to work great except for "xm save/migrate" which fails with
> > > > "Unable to get platform info.: 9" at XendCheckpoint:227.
> > > >
> > > > Playing around with it I found that there is a program called xc_save
> > > > which is called from XendCheckpoint.py and passed in a file handle for
> > > > the xend session and the target file named on the command line. These
> > > > handles seem to be closed when xc_save exec's, so subsequent calls into
> > > > Xen by xc_save fail. I got it to work by adding...
> > > >
> > > >  from fcntl import ioctl;
> > > >  FIONCLEX = 0x5450
> > > >
> > > > near the top of XendCheckpoint.py
> > > >
> > > > and...
> > > >
> > > >  ioctl (fd, FIONCLEX, 0)
> > > >  ioctl (xc.handle(), FIONCLEX, 0)
> > > >
> > > > at the beginning of forkHelper.
> > > >        
> This is clearing the close-on-exec flag for each of the file descriptors, so
>
> import fcntl
> ...
> fcntl.fcntl(fd,          fcntl.F_SETFD, 0)
> fcntl.fcntl(xc.handle(), fcntl.F_SETFD, 0)
>
> ought to do exactly the same, but be a little neater.
>
> What interests me though, is that these flags are set in the first place.  The
> default is for this flag to be unset, and Xend is not setting the flag
> explicitly on either of those file descriptors, so how did they get set?
>
> Could you investigate?
>
> Thanks,
>
> Ewan.
>  
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel