> > then i think it's appropriate to introduce a small lock in ufs inode.
> > protecting it by VOP_LOCK is difficult because of
> > the locking order constraint you mentioned.
> 
> preferably we can find an existing lock to use for this rather than
> adding a new one.  I haven't yet looked enough at this part of the
> vnode locking scheme to say whether any of them would be satisfactory,
> though.

i personally prefer a lock in ufs inode because what it protects is
ufs internal things.
but, if you want to use exisiting locks, i think v_interlock is fine.

YAMAMOTO Takashi