How can you get an incomplete ARP?

Perhaps someone can put this into perspective for me.
This is an age old question that has arisen many times throughout the
years and my web searches have failed to answer this.
The host is visible in arp and shows incomplete, the host cannot be
pinged.
Sometimes the prob is a cable, sometimes a host route setting,
sometimes an intermediary device.
(Let me know if there are other known things that cause this).

ca-santa-barbara-router#sh arp | i 70.169.191
Internet 71.169.191.209 0 Incomplete ARPA
Internet 71.169.191.218 - 0018.731f.407d ARPA Ethernet1

My question is, how is it, that we get the IP address from the
directly connected host into the ARP cache, but not the MAC address?
We arp out for that IP, and the connected host is smart enough to
reply that it has that IP, but it isn't smart enough to send it MAC
address?

Thanks,
crzzy1

 

No, this just means the cisco device has sent an ARP for the IP
in the list (71.169.191.209) and it has not received a reply.

So the host is down or not connected.

 

That is not correct. I forgot to mention that this is a /24,
and only the host 71.169.191.209 is in fact connected.
I can ping the broadcast address, and only this IP comes into the ARP
table.
In this case there was a problem with the routing on the host.
So in other words, my router sees the host, the host in some way
replies that it has this IP, but is unable to say what its MAC is.
I have seen this numerous times, but never have figured out what
really takes place to let my router know that that host is there, but
not what is the MAC?

crzzy1

 

- the nearest I can get to an identification!

Interesting. You have managed to work out how to have some contact
with an interface on a LAN - albeit a not very productive one given
that it relies on a response to a LAN broadcast - *without* getting a
complete entry into the ARP cache.

I believe when an IP node detects that the destination IP address
corresponds to the broadcast address for an address range assigned to
an attached LAN (as determined by the subnet mask), it has no need of
ARP support since it can use the broadcast MAC address on the LAN. I
hope someone who is a practical specialist in these matters rather
than only a theoretical amateur like myself can confirm or deny this
assertion - and, if deny, perhaps provide an alternative explanation.

I think you need to read and understand the ARP RFC which is quite
short, 10 pages, and has a very helpful schematic of the ARP logic.

It may be best to access the RFC, 826, via the Wikipedia article:

http://en.wikipedia.org/wiki/Address_Resolution_Protocol

I see that the article references a 3-page PDF of the logic over which
you can pour.

A quick scan of the RFC indicates that it is not required that the
node sending an ARP request places the requested IP address in the ARP
cache at this time *without*, obviously, the MAC address. However I
can see that some implementations might do this. Actually I could be
wrong - and, if so, I hope for correction.

According to the RFC, this is an impossibility. How are you certain
that the ARP request was sent? If an ARP reply is received from the
interface of another node on the LAN - even when the intended
recipient is not the receiving node - a *complete* ARP entry should be
built for the sending node.

received a reply.

Rob's response indicates that the presumed Cisco implementation logic
bothers to create incomplete ARP entries. I guess this may have the
benefit - assuming it truly is optional - of revealing possible
problems.

It's not clear what you may mean here. If it is a reference to ARP
processing, you will see that, in effect, all ARP processing is "/32",
that is, it concerns only the IP address of individual interfaces
rather than address ranges. As I suggested above, if the destination
IP address happens to correspond to the broadcast IP address of the
address range assigned to the LAN to which the interface about to send
a packet is attached, a broadcast MAC address can be used.

Having been told that the address range is determined by 24 contiguous
bits, I can suppose that the broadcast address you used in the
successful PING command was 71.169.191.255.

If there is only one interface other than the sending interface, IP
address 71.169.191.218 I guess, connected to the LAN, you should
receive responses only from that node - but based on having used the
broadcast MAC address at the Ethernet level.

I think I see at what you have been getting all this time! Is this one
of those famous Cisco extensions to the common interpretation of RFCs?
Could it be that the Cisco implementation of ARP involves the ARP
logic being informed of the IP addresses of interfaces known to fit
the address range assigned to a LAN to which a local interface is
attached in case broadcast requests were used to discover them? It
would appear that this happens at a level in the logic where the MAC
address associated with the IP address has been lost.

I wonder if this mightn't be some way of operating dynamic discovery
of nodes within the network as extensively as possible. I guess one
can speculate endlessly ...

Chris Mason

 

Note:

A host typically creates a temporarey entry in the ARP cache, sends the
ARP request (broadcast) and when/if a reply is received, it then updates
the record in the arp cache with the received ethernet address.

Generally, this is quick enough that you don't see it happening.

 

That is what I said, but he does not believe it.

 

Here is an idea.

"I can ping the broadcast address, and only this IP
comes into the ARP table. "

ca-santa-barbara-router#sh arp | i 70.169.191
Internet 71.169.191.209 0 Incomplete ARPA
Internet 71.169.191.218 - 0018.731f.407d ARPA
Ethernet1

The broadcast ping from 218 just goes out with no ARPing.
In order to reply (with a unicast) the "target" host must
ARP for the ping sender 71.169.191.218.

The router receives the ARP request from 209:-
It replies to it and also does gratuitous ARP or ARP
snooping processing creating the Incomplete entry. For
some reason the entry cannot be completed. Perhaps
after the snooping or gratuiting the router does a real arp
to complete the process and the 209 host does not reply.

I am not sure of the details of gratuitous arp or snooping
so I am not sure as to the plausibility of this hypothesis.

Have you checked that you do not have a subnet mask
mismatch? Well, looking at the addresses I don't
suppose that is possible without a discontiguous mask
which is no longer permitted. Worth a check anyway.

Otherwise, maybe the ARP entry is nothing to do with
your ping and the 209 host is sending some other traffic.
Windows hosts for example are very chatty.

To investigate further you could try:-
- Packet capture with some external device
say with wireshark - hub or SPAN port needed.
- debug arp
- If you have very recent IOS, packet capture on router.
- Packet capture on target (209).
- Check you have no ACLs that could block the ARP.

You might post the router config.

By the way, most people mangle IP addresses
in usenet messages so as to preclude identification.
e.g change the first octet. Maybe you did that already:?)

 

Thanks for the reply.
So are you saying that I am getting an arp reply that doesn't have a
MAC address in it?
I ask this because I am only getting an IP and not an "ethernet
address" that you assert that I am getting in your answer.
My question is how does it get a reply with only an IP and no MAC?
Or if there is no reply, then how does it know that only that single
host out of 254 possibility's is out there?

Thanks,
CJ

 

Thank you for your well worded response.
I did mangle my IP like you mentioned,, yes it would be a breach of
protocol not to.
I think you are correct in your assertion of the gratuitous arp.
This issue was fixed by having the customer concentrate on his routing
on his host side. (I have no visibility into his side, so I couldn't
use wireshark. but I would like to try the packet capture on the
router side next time one of these crop up.
Again though, a really excellent answer.

Thanks,
Crzzy1

 

No, I think you are getting no ARP reply at all.

The "incomplete" entry with only IP in it is created when the router
wants to send something to that IP. The system does not need to exist
for that.

I think it doesn't.

Maybe it has heard traffic from that address.

 

Typically the reverse. It means you have sent an Arp request, but not
gotten any response.

"will IP address X please stand up and reply to me with their ethernet
address ?

BUT, nobody responded.

If this is aon a cisco box, you can:

clear arp <ip address>

This will remove the incomplete arp entry.

you can show arp to confirm.

You can then try to ping that host and you will ether get an incomplete
arp entry or a completed one.

(instead of "ping" you can telnet or any other IP level command that
would try to send an IP packet to that host.


If 10.0.0.2 sends an ARP to ask about 10.0.0.13, then 10.0.0.13 will
implicetely know 10.0.0.2's ethernet address and add that record to its
arp table (allowing it to reply to 10.0.0.2's requests from now on).

10.0.0.13 will then send a response to 10.0.0.2, and 10.0.0.2 will then
get the ethernet address corresponding to 10.0.0.13 and complete the
incomplete arp entry.

Note that you could theoretically have a misbehaving machine on your LAN
which uses a blank or otherwise invalid ethernet address so that when it
responds to arp requests, the responses are considered illegal and not
added to the arp tables (leaving those incomplete entries).

 

Question:

If host1 thinks host2 is part of the same subnet, it will send an arp
asking for host2's ethernet address.

That broadcast arp will specify host2's ip address.

So at the low level, when host2 gets the ethernet broadcast, it will see
"this is something that concerns me" and pass it upwards for processing.

If the ethernet broadcast contained an arp request for an ip address not
used by host2, it would not be passed upwards for processing.

So, by the time the "upwards" layer gets the arp request, does it
blindly respond to the ethernet address of the sender of the broadcast ?

Or does the arp response behave as an IP packet and gets routed to a
router if the requestor is not in the same subnet ?

(at which point the arp request would never get to the requestor,
leaving a incomplete arp record at the requestor's arp database)

 

A few people have tried to explain this. Basically you are not "getting the IP address from the directly connected host" By pinging that IP address. You are telling the router that the address exists. It therefore puts that IP address in its ARP cache with an incolmplete value until it receives an ARP reply. If this does not happen (As in your example). It remains as incomplete. If you ping any other address in your IP range you will see all those addresses appear as incomplete as well. So the router does not see the device on its physical LAN. The fact that it tries to ARP for the IP address means that the router is on the same subnet as the IP address. If this was not the case then it would not ARP for it.

Only things I know that could cause this are VLANs, incorrect address or mask on client or physical network faults, VLAN or MAC ACLs.

I don't know of any reasons why a client would simply not reply to an ARP.