Hi.  I encountering some problem with a very elusive cause, which may be something to do with the WET mode.
I'd like to solicite other testers to see if they can reproduce this.

I'm using experimental CVS. Here is my configuration:

"A" and "B" are WRT54G's.
"C" is a Linux PC (Ubuntu, for all that matters)
"D" is a FreeBSD 5.3-release PC.

C and D are connected to B.
B has a WET bridge to A which is my NAT router to the Internet.
So (you may need monospace fonts to see this diagram)

internet (pppoe) === A ----wireless------        B === ethernet ===  C
                                                                     === ethernet ===  D

Consider the following scenarios

(1) Suppose, from C, I ssh into A. In that ssh session on A, I execute ping D. It fails.
(100% packet loss, details later [*])

(2) Suppose, from D, I ssh into A. In that ssh session on A, I execute ping C. It fails.

(3) Same as (1), but I ping C instead. It works --- I get all the ping replies.
(4) Same as (2), but I ping D instead. It works.

(5) Pinging C directly from D and vice versa of course works.

(6) Most other operations (e.g. Web browsing) from C (or D) to A seem to work fine.
(A is my NAT router to my ISP.)

[*] The ping fails even with all firewalls turned off on all sides,
so it does not seem to be a firewall problem. Also, _sometimes_, when I invoke ping,
I get the ping replies correctly for the first 10 to 20 packets or so, and then the
ping replies start coming in for no reason (without me exiting the ping program).
During all this time, if I ping A from C (or D) that works,
but (1) (or (2)) simultaneously fails.

After some sleuthing, I found out that the problem seems to disappear
if I *do not* use the WET mode, but use WDS instead to connect between A and B.

Finally if I connect C and D directly to A, there is no problem at all.

Would someone kindly try to reproduce this ?

Thanks very much in advance!

// Steve

It doesn't surprise me if you didn't break the bridge.

If you don't break the bridge by assigning different subnets to the client mode router and the AP, you can't expect multiple machines behind the client mode router to work reliablly.

Sounds like the same scenario that was described in this thread:

http://openwrt.org/forum/viewtopic.php?id=1078&p=2

See my reply to that thread (second page); it touches on the reasons why this won't work.

-- Nathan

Ah that explains it
(all along I assumed it was just magic that WET even worked ,
and it half-works with even APs that don't say they explicitly support bridging of any kind).

But since WET is default on OpenWRT (when not running as client)
I think there should be huge warning signs about it not being reliable

Thanks very much for everyone's help,

// Steve

(I'm happily using WDS now, although I'm comtemplating just using layer-3 routing
to get around these problems when I have to connect to an AP that does not support WDS.
Any one have a rough idea how fast a WRT can do layer-3 forwarding?)