Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!uw-beaver!rice!sun-spots-request
From: Neil%Teleos.com@ai.sri.com (Neil Hunt)
Newsgroups: comp.sys.sun
Subject: FIONREAD on datagram internet sockets?
Keywords: SunOS
Message-ID: <8904050232.AA06898@wayback.teleos.com>
Date: 24 Apr 89 20:01:21 GMT
Sender: usenet@rice.edu
Organization: Sun-Spots
Lines: 34
Approved: Sun-Spots@rice.edu
Original-Date: Tue, 4 Apr 89 19:32:57 PDT
X-Sun-Spots-Digest: Volume 7, Issue 244, message 5 of 10

Does anyone know anything about the missing 16 bytes are which are
reported by ioctl(sock, FIONREAD, &n) when applied to an internet datagram
socket?  I find that if ioctl reports N bytes available on the socket,
then a read with a much larger buffer returns a maximum of N-16 bytes.  If
there are multiple datagrams, with lengths N0, N1, ...Nn, then N = N0 + N1
+ ... + Nn + 16, and after packet n has been read, (theoretically leaving
16 bytes, according to the original report) the next read blocks, and an
additional ioctl(sock, FIONREAD, &n) performed at that point reports 0
bytes available.

Here is the code, which is supposed never to block:
(Error checks removed for clarity.)

read_last(sock, packet, len)
char *packet;
{
	ioctl(sock, FIONREAD, &n);

	while(n > 0)
	{
		i = read(sock, packet, len);
		n -= i;
	}

	return i;
}

A workaround is easy.  I was curious whether this is a bug, or if it is
just a case of not finding the right page in the manual (is there such a
thing as a right page in ``Socket Based IPC Tutorial''?)

PS: The system in question is a Sun 2/120, 3.5; also Sun 2/50, 3.5.

Neil/.