On Wed, Apr 11, 2018 at 12:34:03PM +0200, Martin Pieuchot wrote:
> On 09/04/18(Mon) 16:10, Martin Pieuchot wrote:
> > Diff below implements most of the missing locking goo for NFS without
> > enabling it.  It does the following:
> > 
> > - Add a missing PDIRUNLOCK in nfs_lookup()
> > 
> > - Change vrele(9) into vput(9) where necessary.  nfs_nget() will in
> >   future return a locked NFSnode.  Just like ffs_vget() returns a
> >   locked vnode.  So every time it is called we need a corresponding
> >   vput(9).
> > 
> > - Make sure VN_KNOTE() is always called with a valid reference, before
> >   vrele(9) or vput(9).
> > 
> > - Substitute an XXX by an assert in nfs_removeit().
> > 
> > However as long as nfs_lock/unlock are noops, this diff should not
> > introduce any change in behavior.  But please do test this diff :)
> 
> New version of the diff after a review from visa@.  It includes:
> 
> - Stop calling vput(9) inside nfs_mknodrpc() to be able to call
>   VN_KNOTE() with a proper reference in nfs_mknod().
> 
> - Do not unconditionally unlock a node in an error case in
>   nfs_lookitup().
> 
> - Fix a dangling reference in the case of 'goto again' in nfs_create().

OK visa@, with one more tweak:

The KASSERT(VOP_ISLOCKED(sp->s_dvp)) in nfs_removeit() should be
commented out until the locking is enabled. Currently, VOP_ISLOCKED()
returns zero with NFS vnodes.

Reply via email to