Re: [RFC][PATCH 1/8] namespace: use global namespace semaphore

[raven@themaw.net]

On Fri, 3 Jun 2005, Mike Waychison wrote:

> Ian Kent wrote:
> > On Tue, 31 May 2005, Miklos Szeredi wrote:
> > 
> > 
> > > >There's not really anything magic about using the follow link method.
> > > >It's just a bit of a cheat as a directory can't also be a symlink and the
> > > >callback comes at exactly the right time to make stuff happen.
> > >
> > >So is it something like this?
> > >
> > >  - autofs is mounted over direct mountpoint 'dm'
> > >
> > >  - when dm is looked up, autofs's follow_link method is called
> > >
> > >  - this triggers the actual mount in the daemon, which mounts over
> > >    the autofs filesystem (with some trick to avoid recursion)
> > 
> > 
> > Admit, I haven't paid enough attention to recursion.
> > Ouch! Caught by that the other day.
> > 
> > 
> > >  - when the daemon is finished, the autofs's follow_link method
> > >    enters the new mount (how?) and returns
> > 
> > 
> > A function functionally similar to follow_mount of namei.c. Which needs the
> > appropriate vfsmount.
> > 
> > 
> > >
> > >Why do you need the mount topology for this?
> > 
> > 
> > Don't. Actually need the nameidata upto date when the follow_link method is
> > called.
> > 
> 
> However you can't rely on this for expiry checks, as the vfsmount can change
> (--rbind, CLONE_NEWNS).

No. This is a limitation of the old autofs design and will probably 
remain so for the forseable future. It's one of the issues that I hope 
autofsng will address.

I can deal with expirey otherwise.

Ian