On Jun 23, 2006, at 10:26 AM, Bill Studenmund wrote:

> This is kinda what we are doing now, and what I think is a good idea.
>
> The name, however, is bad. We already have a "hold" facility on  
> vnodes,
> which is where writes to backing store keep a vnode in memory even  
> if it's
> not otherwise being actively used. So use different names. :-)

At the very least, it's a "transaction".  Sounds vaguely like  
something you'd want for a journaling infrastructure anyway.

That said, I like agree with Gordon that we really should get away  
from locks, etc. being handled at the VFS layer, and adopting a more  
Sun-like model for it.

-- thorpej