[PATCH] ipv4/ipvs: Convert to kthread API

[Christoph Hellwig]

On Thu, Apr 19, 2007 at 03:59:44PM -0700, Andrew Morton wrote:
> There still seems to be quite a lot of complexity in this driver's
> thread handling which could be removed if we did a full conversion 
> to the kthread API.
> 
> It all looks.... surprisingly complex in there.

It is.  There quite a few interesting oddities in this code:

 - creation of a forker thread.  This is superflous when using the
   kthread infrastructure as a thread created by kthread_create
   always comes from our dedicated forker thread.
 - the infinite retry on failure looks very bogus, the system
   doesn't recover very well if you try to fork forever in a loop :)
 - a lot of very overlapping state variables.  My reading of the
   code suggests that both a 'master' and 'backup' thread can
   run at the same time.  I think the code would benefit a lot
   from totally separating these codepathes.
 - start_sync_thread and stop_sync_thread are called with
   unchecked user supplied arguments and bug if they don't
   match the expected values.  While all this is under
   capable(CAP_NET_ADMIN) it still sounds like something to
   fix.
 - and the usual removal of semaphores and completions for
   startup/shutdown would benefit the code a lot, as for most
   thread users.