[RFC,PATCH 0/4] Dynamic Pseudo Root

[Steve Dickson]

J. Bruce Fields wrote:
> 
> But maybe there's a way we could make this all work with no or minimal
> changes to the export interface.  At a minimum I assume the kernel would
> have to stop looking up the dentry for each new export it learns about,
> as that would require mounting every exported filesystem on nfsd
> startup.  That seems like a potential scalability and start-up time
> problem.
Esp if there is no 4v traffic... Although I do think the approach of 
crossing over mounts from pseudo to real, does work well (due to the fact
the kernel does handle crossover-mounts very seamlessly). But the cost of the 
actual mount (and umount) make this a bit prohibitive... 

Too bad there is not some type of in-kernel mount (or pseudo mount) that sets 
some bit that would allow the nfsd crossover code just work...

> 
> If we go with an in-kernel solution, are there any advantages to using
> the real filesystem(s) over cooking up a fake in-kernel filesystem?
If we go with a new in-core pseudo root we could stay with the mapping
approach which means there would be less impact to the current server
code (assuming we could come up with some pseudo mount scheme)

On the other hand, as Brent points out, using real filesystems
probably would be simpler and more straightforward, although we
would have to sprinkle "if (dentry not exported) continue" check
through out the lookup and readdir paths...     

I've very indecisive... :-\

steved.