On Wed, Nov 29, 2000 at 12:04:47AM -0700, Warner Losh wrote:

 > The race saved by getfh() I think causes kernel resources to be
 > consumed.  The getfh call locks the leaf and leaving it locked on
 > return to userland seems to be dangerous to me.

What on earth are you talking about?  If that were actually true, all
sorts of hell would break loose.

I suggest you actually read the code.  Here, let me quote it for you:

        NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE,
            SCARG(uap, fname), p);
        error = namei(&nd);
        if (error)
                return (error);
        vp = nd.ni_vp;		<--- vnode is locked due to LOCKLEAF
        memset((caddr_t)&fh, 0, sizeof(fh));
        fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
        error = VFS_VPTOFH(vp, &fh.fh_fid);
        vput(vp);		<--- unlocks the vnode

The file handle has a "generation" number in it, which is incremented when
the vnode is recycled ("stale file handle" ring a bell?), so it's safe to
return the file handle to userland with the vnode unlocked.

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>