Re: [9fans] strange behavior with new Wavelan firmware + possible devmnt bug

Re: [9fans] strange behavior with new Wavelan firmware + possible devmnt bug

[Mike Haertel]

>Pb and I downloaded the 8.10 firmware into a card we had and tried it
>on an IBM T23 laptop.  It worked fine, i.e., we could DHCP, import file
>systems and scp to the laptop.
>
>What flavor machine did you try this on?

A Toshiba Satellite Pro 4280.

I never tried scp to the laptop.  scp *from* the laptop hung for me though.

Hmm.  Did you have WEP enabled with a 128 bit key?

re: [9fans] strange behavior with new Wavelan firmware + possible devmnt bug

[presotto]

Pb and I downloaded the 8.10 firmware into a card we had and tried it
on an IBM T23 laptop.  It worked fine, i.e., we could DHCP, import file
systems and scp to the laptop.

What flavor machine did you try this on?

[9fans] strange behavior with new Wavelan firmware + possible devmnt bug

[Mike Haertel]

The 8.10 firmware for Wavelan cards purports to fix WEP security holes.
However, I just discovered that the 8.10 firmware does not work with Plan 9.

It appears to work a little--ipconfig succeeds in getting an IP address
from a DHCP server.

Interactive ssh connections to a remote host work for awhile but then
get hung.  However "scp filename remotehost:" gets hung immediately.

Theory: the problem is large packets.

So, i tried:
term% echo -n mtu 576 > /net/ipifc/0/ctl

Now ssh logins work, and scp works.

However, attempts to run replica/pull from the sources.cs.bell-labs.com
repository don't work.  On further investigation, I found that it hangs
when attempting to cmp the last 1024 bytes of dist/replica/plan9.log.
Once again we're looking at large packets.  But in this case I had
already shrunk the MTU and other tcp services worked fine.

So there appear to be 2 problems here:

1.  Either the Wavelan driver is broken, or the 8.10 firmware is broken.
I am inclined to believe the firmware is at fault.  I downgraded to the
7.52 firmware and am having no trouble.

2.  devmnt appears to be not interacting well with network interfaces
configured with a smaller-than-normal MTU.  I confirmed this by trying
the same replica/pull, using the old 7.52 wavelan firmware, but setting
the mtu to 576 bytes.  Identical hang when attempting to cmp plan9.log,
so that problem is independent of the firmware version and depends only
on the MTU.