On Jul 22, 2005, at 4:15 PM, Quentin Garnier wrote:

> I might not be understanding the idea correctly, but I don't think the
> <fs>_args structure can be mapped into an uio in a reasonable way.

No different than passing a flat buffer, except you can tag it as  
SYSSPACE or USERSPACE.

> The most evil fs I could find so far is umap.  struct umap_args  
> contains
> two pointers to arrays of u_long couples.

I would actually love to get rid of the stupid binary argument stuff  
for mount, and go with a strings-based approach.

> Packing it up into a uio would mean sys_mount itself has to know about
> how umapfs store they mount data, which kind of defeat the ability
> for sys_mount to simply call VFS_MOUNT() for the mounted device and be
> done with it.

How is that any different than "translating it and then passing  
MNT_KERNSPACE" (or whatever)?

-- thorpej