des@ reported a panic in the NFSv4 server, where
nfsv4_fillattr() did a VOP_PATHCONF() without having
"vp" locked. kib@ suggested D51398 to fix it.

This patch is similar, but puts the NFSVOPLOCK()
call in one place near the top of the function.

des@, please test.

I will be building a kernel with DEBUG_VFS_LOCKS
so that I can find out if there are any more of these
bugs. (Yes, I have just been building GENERIC. Bad me.)