Re: [PATCH] VFS autmounter support v2

[viro@parcelfarce.linux.theplanet.co.uk]

On Thu, Jun 19, 2003 at 11:13:42AM -0400, Mike Waychison wrote:
 
> Introducing special trap vfsmounts w/o super_blocks means we can no 
> longer have arbitrary actions on those traps.  AFS wants to define what 
> happens in kernelspace, autofs wants to define it in userspace.  Last I 
> checked, vfsmount doesn't have an ops structure.

It would have send an event over attached opened file.  Attached at
creation time.
 
> This only works for mounts performed in kernel space.  It doesn't lend 
> itself to performing mounts in userspace and would force autofs to 
> re-implement mount(1) parsing/struct packing in kernelspace.  Definitely 
> not a good solution.

Or if passed event contains opened mountpoint-to-be.
 
> I'm still partial to the idea that a usenamespace ioctl on 
> /proc/<pid>/mounts is a cleaner solution in the long run, both for 
> automounting as well as for administration tools.

Vetoed.  ioctl() is _not_ an acceptable way to implement any generic
functionality.  It basically says "my interface is a garbage".

And yes, we need to think about a new syscall for mount-related
work.  With sane API - mount(2) one is _not_.  sys_mount() would
still stay, obviously.