On Fri, Jun 23, 2006 at 04:37:57PM +0000, Christos Zoulas wrote:
> >But it does not work when you are not using Xen! jeez, can we please
> >test things?
> 
> And by simple code inspection:
> 
> 1. If you don't kernfs_alloctype (which obviously xen does so it works)
>    the splay tree will be empty so the read method will not be found.

Yes, that's what I found. All the ops using kernfs_try_fileop() are broken.
I can confirm that write are broken too, and the write code has been there
for years.

> 2. kernfs_try_fileop() is being passed an errno of 0 in the read or write
>    case, so if the fileop is not found the operation will silently succeed.
>    I don't think that this is how the read case was handled before. I think
>    that they should all return EOPNOTSUPP or something in the case the
>    fileop is not found.

Agreed for read(). write in this state for years, so I don't know if
write silently failing was intended or not.

> 
> So all the ports are broken except of xen. Either back it out again,
> or fix it ASAP.

This code has been there for more than 3 months. Either kernfs is not
used much outside of Xen and leaving it in this state a few more days while
a real fix is found is not a big deal, it broke recently and my code is 
not to blame. 

Anyway I'm looking at it.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--