[PATCH 0/5] Dynamic Pseudo Root

[Steve Dickson]

Chuck Lever wrote:
> On Feb 18, 2008, at 7:32 AM, Steve Dickson wrote:
>> Comments?
> 
> A couple of broad questions:
> 
> What happens if the kernel doesn't have support for tmpfs?  Do you
> provide a facility for falling back to a permanent file system type?
No. If tmpfs does not exist or is not configured, the -R option 
will not work.

> 
> Can you compare your proposal to implementing the bind mounts and tmpfs
> pseudoroot in the NFS server start up scripts?  I'm curious if you doing
> much that requires specific and special support in mountd, etc. or is
> this just writing C code that could be implemented via a start-up
> script?  I like the idea of a user-space implementation, and given the
> relative simplicity of your proposal, creating a pseudoroot via shell
> scripting might be more flexible and less code.
hmm... Interesting concept... off the top, the first problem I see
is the exports have to exist before the --bind mounts. Meaning 
exportfs -a has to happen to populate /var/lib/nfs/etab. Plus
how would re-exports work (i.e. exportfs -arv)? 

> 
> I don't see anything that adjusts the pseudoroot bind mounts if an
> administrator comes by later and adds more exports or takes them away. 
> Is anything needed in "exportfs" to hook up new bind mounts, or remove
> ones that are going away?  Using an automounter here might be appropriate.
No. All the code is contained to rpc.mountd. When exportfs re-exports things
(ala exportfs -arv), the entire v4root tree is torn down and rebuilt.  


steved.