Out of band messages TCP_IP

[Chris Torek]

In article <2773 at servax0.essex.ac.uk> peter at sersun0.essex.ac.uk (Allott P)
writes:
>We have "discovered" an apparent limit of 1 byte for the length of OUT OF
>BAND messages.

(What, for loud out of band messages? :-) )

>We are using Sun-os 4, but have the strong feeling that this is much
>more general.

It certainly is.

TCP does not *have* out of band messages.  It has, instead, `urgent
data'.  Those who implemented 4.2BSD (you know who you are :-) ) chose
to treat the first byte of a region marked `urgent' as `out of band'
data.  It is not, really---in fact, there is no way to tell exactly
which bytes the urgent pointer really points to---but this
interpretation is as wide-spread as 4.2BSD and TCP systems derived
therefrom.  (Almost all Unix TCP implementations are 4.2BSD-based, and
even many VMS TCPs have the same ultimate origin.)

XNS *does* have out of band messages.  They are one byte long.  Network
protocols that have out of band messages typically restrict them to one
(or a few) bytes.  This, of course, is the reason for the 4BSD
interpretation.

>... suggestions of work rounds would be appreciated, we need a method of
>reporting fatal errors over an active link.

(a) use a separate socket; (b) report failure in-band (wrap a protocol on
top of TCP).
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain:	chris at cs.umd.edu	Path:	uunet!mimsy!chris