Pass a fd to program in another mount namespace?

Pass a fd to program in another mount namespace?

[Rob van der Hoeven]

Hi,

I was wondering if it is possible to send a file descriptor to a program
running in another mount namespace. Sendmsg/recvmsg can pass fd's in the
ancillary data and this works great until my program switches to another
mount namespace using:

unshare(CLONE_NEWNS)
pivot_root(".", "./us_rootfs")

If there is a way to pass descriptors across mount namespaces then it
would greatly benefit my program so I hope this is possible. 

Best regards,
Rob van der Hoeven.
http://freedomboxblog.nl

Re: Pass a fd to program in another mount namespace?

[Serge Hallyn]

Quoting Rob van der Hoeven (robvanderhoeven-nv30jWqA7Mk@public.gmane.org):
> Hi,
> 
> I was wondering if it is possible to send a file descriptor to a program
> running in another mount namespace. Sendmsg/recvmsg can pass fd's in the
> ancillary data and this works great until my program switches to another
> mount namespace using:
> 
> unshare(CLONE_NEWNS)
> pivot_root(".", "./us_rootfs")
> 
> If there is a way to pass descriptors across mount namespaces then it
> would greatly benefit my program so I hope this is possible. 

Yes, it is possible.  Your email makes it sound like you've tried
and it didn't work for you, in which case a short test case showing
what fails might be useful here.

-serge

Re: Pass a fd to program in another mount namespace?

[Rob van der Hoeven]

On Tue, 2014-01-21 at 17:07 -0600, Serge Hallyn wrote:
> Quoting Rob van der Hoeven (robvanderhoeven-nv30jWqA7Mk@public.gmane.org):
> > Hi,
> > 
> > I was wondering if it is possible to send a file descriptor to a program
> > running in another mount namespace. Sendmsg/recvmsg can pass fd's in the
> > ancillary data and this works great until my program switches to another
> > mount namespace using:
> > 
> > unshare(CLONE_NEWNS)
> > pivot_root(".", "./us_rootfs")
> > 
> > If there is a way to pass descriptors across mount namespaces then it
> > would greatly benefit my program so I hope this is possible. 
> 
> Yes, it is possible.  Your email makes it sound like you've tried
> and it didn't work for you, in which case a short test case showing
> what fails might be useful here.
> 
> -serge

Hi Serge,

Problem solved. It turned out that I had not properly initialized the
ancillary header. Because the problem only showed itself after switching
to another mount namespace I wrongly assumed that this had something to
do with the error. Very happy that this works now.

Your answer made me investigate this further, thanks!
Rob van der Hoeven.
http://freedomboxblog.nl