[comp.protocols.tcp-ip] Paranoia: fragmentation of UDP packets at user level?

tpc@bnr-fos.UUCP (Tom Chmara) (05/18/89)

I'm writing YAMI (Yet Another Message Interface) to use unconnected datagram
sockets on a SUN (OS 4.0.1) workstation.
	I suddenly realized that I wasn't sure whether a UDP packet HAD to
be returned in its entirety in response to a recv() or read() call.
That is, is there any possibility that a read() of a UDP packet could result
in a --partial-- read being returned?
	I *think* UDP is all-or-nothing (witness the checksum):  if you get it,
you get ALL of it; otherwise, you don't get it.  (you can, of course, get
clones..  but that's another issue).  RFC 768 suggests this type of interface
with IP, but it doesn't seem to require it.
	I sure hope I'm right; I'd hate to have to figure out a way of
regaining synchronization on a UDP channel...

Thanks for your time...
	---tpc---

-- 
I am sole owner of the above opinions. Licensing inquiries welcome.
------------------------------------------------------------------------
Tom Chmara			UUCP:  ..utgpu!bnr-vpa!bnr-fos!tpc
BNR Ltd.  			BITNET: TPC@BNR.CA