Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!apple!agate!ucbvax!TWG.COM!ljm
From: ljm@TWG.COM (Leo J McLaughlin)
Newsgroups: comp.protocols.tcp-ip
Subject: NetBIOS over TCP/IP
Message-ID: <8912211035.aa20407@Obelix.TWG.COM>
Date: 21 Dec 89 18:18:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 34


There appears to be a difficulty in NetBIOS over TCP/IP's handling
of simultaneous, duplicate name registration requests.  If two nodes
each ask for the same name at roughly the same time (for example if
they are both coming up from the same power failure) both nodes will
believe they own the name.

This expected behaviour results from the stipulation in section 5.1.1.1
of RFC 1002, B-NODE ADD NAME, that names are not added to the local table
until after the NAME REGISTRATION REQUESTS and NAME UPDATE REQUEST are sent
and RFC 1002 section 5.1.1.5, B-NODE INCOMING PACKET PROCESSING, which
requires that incoming NAME REGISTRATION REQUESTS be ignored until the name
is added to the local name table.  Furthermore, section 5.1.1.5 states that
incoming NAME UPDATE REQUESTS have no effect other than to update the
optional IP address cache.

This issue is further clouded by a confusion within RFCs 1001 and 1002
about the name of this final name registration packet.  Section 15.2.1
of RFC 1001, NAME REGISTRATION BY B NODES, shows a NAME OVERWRITE DEMAND
rather than a NAME UPDATE REQUEST and Section 4.2 of RFC 1002, NAME SERVICE
PACKETS, describes the existance of a NAME OVERWRITE DEMAND packet but no
NAME UPDATE REQUEST packet, however, section 5.1.1 of RFC 1002, B-NODE
ACTIVITY, the actual NetBIOS over TCP/IP B-node protocol specification,
never mentions NAME OVERWRITE DEMANDS just NAME UPDATE REQUESTS.

Note that the MAP/TOP NetBIOS over OSI specification avoids this difficulty
by defining a 'being registered state' during which the name is defended
from other name registration requests but doesn't answer to name query
requests.

If any implementors of NetBIOS over TCP/IP have dealt with this issue
I would be interested in knowing if they came to the same conclusion about
the NetBIOS specification and if they decided to implement name resolution
as per RFC or if they fixed the problem.