Re: WAPBL patch for testing

[Thor Lancelot Simon <tls%rek.tjls.com@localhost>]

On Sat, Nov 01, 2008 at 10:28:14AM +0100, Joerg Sonnenberger wrote:
> On Sat, Nov 01, 2008 at 09:47:24AM +0100, Juergen Hannken-Illjes wrote:
> > One thread has a wapbl transaction, calls VOP_GETPAGES and wants
> > to busy a page  while  another thread calls VOP_PUTPAGES on the
> > same vnode, takes pages busy and wants to start a wapbl transaction.
> 
> Does that involve snapshots or not? Without snapshots, genfs only has to
> take the wapbl lock in the case of GOP_ALLOC in VOP_GETPAGES. If you can
> get it in this situation, please provide the full stack traces of the
> involved lwps. For snapshots, the real problem is that the strategy of
> fss doesn't handle WAPBL as it should.  It seems to somewhat work, but I
> can't stress it much.

Um.  The snapshot code was there first.  I'd say it's considerably more
accurate to call the problem "snapshots were not adequately considered
when WAPBL was ported".

I know there was an urgent need for WAPBL but I keep looking at the code
and it really is somewhat irritating to see a second kind of "transaction"
pop up after Juergen did so much work on fstrans -- this is sort of an
impedeance-matching problem between the two notions of what's a transaction
and how you start and end one, a situation that isn't very likely to get
better without a lot of hacks.

Thor