[jcifs] no name with type 0x00 with no scope for host

Christopher R.Hertel crh at ubiqx.org
Thu Feb 21 07:44:03 EST 2002


On Wed, Feb 20, 2002 at 02:50:25PM -0500, Michael B Allen wrote:
> On Wed, 20 Feb 2002 11:44:05 -0600
> Christopher R.Hertel <crh at ubiqx.org> wrote:
> 
> > By specifying that "no name with type 0x00" was found, it suggests that
> > the query was specifically looking for a name with that suffix, and that
> > it may have missed other names.
> > 
> 
> It *is* looking for a name with that suffix but I don't see why you think
> it could miss the other names;

I don't think that the NODE STATUS REQUEST will miss names with that
suffix, I think that the error message is confusing.

> they're all returned in the node status.  If
> you get a response to the node status at all, the name with that hexCode
> (in this case <00>) is being advertised and the getAllByAddress call
> is ultimately successfull. If you don't get a response it doesn't know
> about <00> so we throw what I think is a very informative and accurate
> UnknownHostException.

I find it inaccurate.  It's like saying "no name 'foo' was found" when we
were not looking for 'foo'.  We were looking for any name, with any
suffix.  Saying no name with type 0x00 was found is the same kind of
problem, in my mind.

[I should mention that the Scope ID does restrict the NODE STATUS REQUEST.
There is no wildcard scope ID.  If you specify a Scope ID of "", then you
should not get a response from a node in scope "banana", and vice versa.]

> You're thinking "get node status" when you need to be thinking "get all
> the names at the host advertising this name".

Right.  That is what I'm thinking.  The "get node status" uses the
wildcard name which, since it is a wildcard, doesn't really have a name
type.  It's *any* name, no matter what the suffix.

> Let's go one level down to see how jCIFS does the getAllByAddress
> exactly. Theres not a lot of code behind this BTW so looking at the
> source is the best way but here goes.

Have (just a short while ago).  You're right, of course, that that's where
the answers are.

> When you call:
> 
>   getAllByAddress( NbtAddress addr )
> 
> it calls:
> 
>   addrs[] client.getNodeStatus( addr )
> 
> where client is a static instace of NameServiceClient and addrs is
> the resulting array of NbtAddress' representing the names advertised
> by the host at addr. Every NbtAddress has an IP and a Name object as
> a member. Every Name object must have a String name, and int hexCode,
> and a String scope but the scope can be null.
> 
> Now getNodeStatus builds a NODE_STATUS_RESPONSE message, encodes it in
> a datagram packet and sends it to the IP associated with addr.

I think you mean a NODE_STATUS_REQUEST, yes?

> Here's the pertaintent part; IF you get a response, that name with
> whatever hexCode is being advertised by the target (in practice <00>
> is the only hexCode that works reliably), your getAllByAddress call is
> successfull and the original addr object will be in the array of other
> addrs[]. If however the host is NOT advertising a name with that hexCode,
> you will not get a response and an UnknownHostException: no name with
> type <hexCode> ... is thrown.
> 
> In Ashish's case, he called:
> 
>   getAllByAddress( String host )
> 
> which is just shorthand for:
> 
>   getAllByAddress( getByName( host, 0x00, null ))
> 
> and if host is a dot quad IP getByName is a noop.

Okay, that's where the problem is, and it goes back to my original
message.  You say above "in practice <00> is the only hexCode that works
reliably"... but it's not since XP doesn't register it by default like the
others do.

I see what is going on in the code above, and it's cool that it can be
done, but it can also be misleading and problematic.

In the getByName(), if the host is not a dot quad IP then you are stuck
(particularly with XP) since you cannot know that an 0x00 name will be
present.  The lack of the 0x00 name doesn't mean that the host doesn't
exist within the NBT network.

If the host in getByName() is a dot quad (or a DNS name or something else
we can resolve) then you'll have an IP address.  You no longer need to
consider the 0x00 suffix.  It is no longer relevant to the query.

> > environment that lead me astray.  I still think that the particular error
> > message is misleading.  If a "real" name and suffix are specified, then
> > I'd want to see "name <name> with suffix <suffix> and scope <scope> was
> > not found".  If the wildcard name is used, then I would like to see "no
> > status response from host <ip address>".  Or something along those lines.
> > 
> 
> The NODE_STATUS_REQUEST message is a type of direct name query. This
> physical layout of the message is exactly the same. The only thing that
> distingushes a NAME_QUERY_REQUEST from a NODE_STATUS_REQUEST is one
> single bit.

Yes.  Well, no...  Three bits.  The B and RD bits are significant in the
NAME QUERY but not in the NODE QUERY.  The bit you're talking about is the
difference between 0x0020 and 0x0021 in the QUESTION_TYPE.  (Sorry, being
pedantic again.)  :)

> The QUESTION_NAME in both messages is a "real name" and
> therefore has a hexCode associated with it. It can be <20>, <1E>, whatever. 
> In practice MS hosts only reliably respond to <00> though. You can try this
> with:
> 
>   getAllByAddress( "ip", 0xXX, null );

I understand what you are doing here, and why you picked the 0x00 suffix.  
The problem is that the 0x00 suffix is only *more* reliable than any
other.  Unfortunately, it is not completely reliable, as XP emphasises.

The RFCs require that a permanent name be registered to avoid exactly this
problem.  Microsoft failed to follow spec.

> > 
> > Of course, we still don't know what's going on in Ashish's case.
> 
> I thought we figured this out? He's running on the wrong interface. I
> directed him to the laddr and baddr properties.

His later messages suggest otherwise.

If he is trying the lookup using a host name, then the code will be:

    getAllByAddress( getByName( host, 0x00, null ) )

and the lookup *will* fail because XP doesn't register the 0x00 name.  In
that case, the message saying no name <host> with type <00> exists is
correct.  This would be the equivalent of doing:

  nbtstat -a <hostname>

The question is, what does Microsoft do in that case?  If they only check
for the <00> name then they, too, would return an Host Not Found error.

If he is using an IP address, then:

a) He should get a response if 'nbtstat -A <ip>' gets a response.  If not,
   I'd want to see a packet trace.

b) Even if he's doing something wrong, my own 2cents is that a status 
   query using an IP address that does not get a response should report
   "no status response from host <ip> was received" rather than "no name
   with type 0x00" was found".

Another reason for this is that some W/95 boxes send query replies to the
wrong port number, so a "no status response from host <ip> was
received" seems more accurate to me.

You know I enjoy hair-splitting discussions like this.  ;)

Chris -)-----

--
Samba Team -- http://www.samba.org/     -)-----   Christopher R. Hertel
jCIFS Team -- http://jcifs.samba.org/   -)-----   ubiqx development, uninq.
ubiqx Team -- http://www.ubiqx.org/     -)-----   crh at ubiqx.mn.org




More information about the jcifs mailing list