Fixing exec-modifies-atime

[Bruce Evans]

On Mon, 7 Feb 2005, Ken Smith wrote:

> Does anyone see any problems with this approach, or any major flaws
> with the implementation in this patch?

I found some new problems (more details in private mail):
- using VOP_SETATTR() makes us wait for snapshots to complete, but we
  shouldn't have to wait for operations that don't really involve
  writing.  The wait may be very long if a snapshot is in progress
  and affect many processes since execs are fairly common (though
  perhaps less common than writes).
- the implementation makes even bogus execs like exec of "/" wait
  for snapshots to complete.  This is not the serious problem that
  I thought it was when I wrote the private mail -- the vnode is
  not locked, so only the process doing the silly exec is affected.

Bruce