Hi guys.

Here's the setup:

PC1 <---> AP1 <----> Client <---> PC2 <---> Local net

PC1, AP1, Client, and client side of PC2 are on the same
net, the other side of PC2 is on the local net.
I've set up all the routing as I thought it should be, and found
that I cannot ping (or anything else) from the 'local net' to PC1.

Two days of headbreaking, till I just randomly tried to telnet from PC1
to Client (which didn't work - client has no Telnet server), but the net
started working immediately. If I wait a while, the net breaks down
and another telnet is in order (or an FTP works too).

This is really beyond me. I'm suspecting ARP problems, mostly because
the 'breakown' time seems to be in the order of the ARP cycle time.
Could it be that the ARP doesn't work in one direction?

I'd really appreciate suggestions here.

John

PS: AP1 and Client are both Edimax 7209 Wifis, one as AP, the other
as Client. PC's run kernels 2.6.20+. No firewalls installed (yet).

You're a bit sketchy on the details ... I can't make any sense of your symptoms, so I suggest to double check the basics:

Do you have normal access from PC2 to PC1?

Did you enable ip_forwarding on PC2?

Do the local net clients have DNS set up appropriately for your layout?

Beyond that, the only thing I know to suggest is to run a packet sniffer such as tcpdump or wiresharl on PC2 and do some direct troubleshooting of what is going on. BTW, tcpdump (and probably wireshark -- I am less familiar with it) can show ARP packets along with the higher level protocols.

Thanks Blackhole54...

Well, here are a few more measurements I did today. First your questions:
- I don't have access from PC2 to PC1, until I do a telnet intent from PC1 to 'Client'.
I can't ping either from PC2 to PC1, until I do that.

- I have ip_forward enabled on all machines. I learned this the hard way some time
ago. (also, after the telnet intent (or ftp), things forward perfectly.

- I have the ISP's DNSs correctly installed in both machines, and DNS starts working
on PC2 after the telnet.

An extra point is that Internet is connected to PC1 (through another nic).

New is that I cannot see any ARP requests from PC2 activity coming through the Wifi link.

A strange thing is that ARP requests from PC1 to 'Client' and PC2 return the same MAC.
This made me suspect that the Client had the MAC cloned from PC2's nic, but I went and
checked that. The MAC in the Client's web interface is set to 00:00:00:00:00:00, which I suppose
means it should use its own MAC (I can't find any option to return it explicitly to its
default value, and the client is quite inaccessible - so a hard reset is difficult).

Could this be the cause of not relaying ARP requests?

John

I am not certain about your topology.

I assume AP1 is probably a wireless access point. It appears both PC1 and client connect to it. Since your are doing ARP requests from PC1 to client, I assume both PC1 and client must be on the same "side" such that AP1 is just acting like a switch for them. What I'm not clear on are the two connections going into client. Is client running a bridge?

You speak of a web interface on client. Just exactly what is client?

Ok... Here's a list:

PC1: Two NICs, one to internet, the other to the AP1 (Access point)
Both are on separate subnets (say 2.x to internet, 9.x to AP)

AP1: Edimax 7209, Access point, net is 9.x

Client1: Edimax 7209, as Client ('structured'), net is 9.x

PC2: Two NICs, one to the Client (net 9.x), the other to the local
net (net 1.x)

I cannot:
- connect or ping from PC1 to PC2 (1.x side)
- I cannot connect (or ping) from any computer on PC2's net (1.x)
to PC1. (I tried ssh to PC2, then ssh to another computer on 1.x,
which works, but then a ping back doesn't).

I _can_
- connect (and ping) from PC1 to PC2 (9.x side)


As commented before, by chance, I detected that trying a telnet from PC1
to Client, (even though there is no actual telnet server there, so it
gets refused), then things start working normally.

Thanks for your attention

I may have bit off more than I can chew here

But focusing back on the fact that both client and PC2 respond to ARP requests from PC1 with the same MAC ... Does this MAC address correspond to what ifconfig shows as PC2's actual MAC? If so, what happens if you change PC2's MAC to something else (different from all other MACs on network!)? I.e. does client start reporting that MAC also? Do things start working?

If the driver supports it, you can tell ifconfigto use a MAC address other than the assigned one.

Ok... Can't test it right now, as the link is down, but I could try it tomorrow.
At least that would clear up if the modem is answering the ARP calls instead of the
actual PC2. ifconfing in PC2 gives the correct MAC of the Netronix NIC. Also, the
arp cache in PC2 shows the correct values.

Things actually got worse today - to get the link working, it is actually necessary
to connect to the web-interface of Client before pings and the lot work.

Also, there seems to be some kind of weird streak in the web-interface. If I connect
using Firefox (or another 'high level' browser) I cannot explicitly choose to clone
the MAC or use the default value. I'm just guessing that putting 00:00... does that.

If I connect to the web interface using 'links', I actually get a more complete
menu where the option is explicit (Clone, power MAC, or a user defined MAC). I'll try
that with Client tomorrow too.

I wonder if there is another GUI than Edimax's to the mostly third party open source
running in the wireless modem.

John

FYI, I've never figured out what a Edimax 7209 is. I've found references to it on the net but nothing like a manual or anything that might help me figure your situation out. Of course, even if I understood the device perfectly I might still come up empty handed.

Hopefully you will learn something userful when you can run the tests. Good luck.

> FYI, I've never figured out what a Edimax 7209 is.

Ok... They call it Wireless Router, I guess. I'm getting somewhat tired of the firmware.

I is RTL8186 based, and runs Linux, so I'll have a look around if I can upload other
firmware to the thing, maybe with a more extensive setup which lets me check all configuration.
It has busybox installed so I suppose I have access to the internals (maybe only during the
first couple of seconds).

Any suggestions for replacement firmware?

Thanks for everything.
John

Definitely outside my experience. I know the OpenWRT project is popular, but glancing at their website, I don't think your hardware is supported (yet). You might want to double check. If you're feeling ambitious and adventurous, remember that the manufacturer must publish the source for any GPL code on the box., so you should be able to modify it. Do remember that flashing the device with any new firmware caries some degree of danger.