My only nit is that if PR_PURGEIF is set, assume pr_usrreq is not NULL.

You missed the gratuitous setting of purged = 1, copied from above.

  You could add an KASSERT to make certain of that in a DIAGNOSTIC kernel.

Adding the pr_usrreq != NULL assumption seems OK, but I have long been
troubled by the lack of explicit documentation of such invariants
(the lack of which I realize is traditional in BSD).  Adding a KASSERT
is a way of documenting the assumption, and I could add a comment
where PR_PURGEIF is defined.