Re: CVS commit: src/sys/ufs/ufs

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Sun, Sep 20, 2009 at 08:23:38PM +0300, Antti Kantee wrote:
> > In ufs_ihashget(), vget() can return a vnode that has been vclean'ed because
> > vget() can sleep. After vget returns, check that vp is still connected with
> > ip, and that ip still points to the inode we want. This fix the NULL
> > pointer dereference in ufs_fhtovp() I've been seeing on a NFS server.
> 
> Um, hold the phone.  The whole point of vget() is to provide race-free
> access to the weak vnode reference held by the file system.  Are you
> saying this does not hold anymore?

It depends on what you mean with "race-free". If you mean that the
vnode returned by vget() can't be recygled, I think this is true.
If you mean that vget() can't return a clean vnode then this is false:
vget() can sleep in vn_lock(), and it releases the v_interlock mutex before
sleeping. While sleeping vclean() can VOP_RECLAIM() the vnode, even
if v_usecount is > 1.

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