In message <87y8n1fi3w.fsf@snark.piermont.com>,
"Perry E. Metzger" writes:

>I really wish we had sendfile(). :(

You mean we still don't?

senfile is what I first developed a kcont(9)-like framework for.  If
we had UMC-style ops to say:

  ``here's a struct buf,
    please wrap the contents in an external mbuf,
    mark the mbuf as immutable,
    make sure writes to the buffer cache dont change it''

then its straightforward to implement sendfile() using that,
well-designed in-kernel POSIX-aio APIs, and kcont(9).

Without the buf-to-mbuf-chain ops, you need another copy.