At 11:25 02/14/96, Jeff Kell wrote:
> Can someone verify my belief that if an active TCP/IP connection should
> receive an ICMP Host Unreachable, the connection will be dropped?
>
> Or is this hidden from the application layer and ignored?
 
It is my understanding (or at least belief) that these errors aren't
_supposed_ to break TCP connections, but it is also my experience that they
do on many implementations. I don't know what the RFCs say. Actually, it
may be that an error is returned on a send(), and the client program panics
instead of continuing.
 
The reason I don't think they _should_ is that IP routing protocols have
been designed to be self-repairing. If a routing instability produces an
ICMP reject during an established connection, I would rather TCP gave me a
little assistance and tried again (and again, in anticipation of the
self-repair).
 
Unfortunately, I have seen connections dropped (VT, rlogin, telnet, ftp)
due to just such errors. It was never convenient and always an irritation.
And usually transient, so I could just reconnect and try to recover what I
had been doing.
 
It's certainly OK if the ICMP rejects prevent connection establishment:
maybe there's no such host!
 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Mike Casteel                      Unison Software, Inc.
[log in to unmask]                    Seattle, WA
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~