Beaglebone software issues [was: Using beaglebones as boards for the ietf "homenet" and "babel" working groups]

GoogleGroups
1

Hi Jason, and thanks for your kind reply.

The larger and more official for .org boards is “beagleboard”,
http://BeagleBoard.org/discuss or beagleboard@googlegroups.com.

I’ve tried to subscribe, but it asks me for a Google account.

(Snipped bunch of helpful info, will look into it.)

The critical bit is to remove connman, which broke the WiFi on a network
with a DHCPv6 server

Grrr connman. It drives us nuts all the time.

Why not simply use the static /etc/network/interfaces config, perhaps with
a script to edit it automatically? Right now, your system has three
different methods for configuring interfaces (/etc/network, connman, and
the hotspot thingie), which is a lot of conceptual load for somebody new
to Linux.

I think Robert would have to reply, but I’m guessing that connman provides a GUI and lots of seemingly useful defaults while remaining relatively small.

Can /etc/network handle it all? How do you enable a GUI or even CUI to edit it?

There is a race condition between booting and loading the Bluetooth
firmware. Reloading the firmware manually works around the issue.

This has been haunting us for a while.

Okay, that’s good to know, I’ll keep hacking at it.

Thanks for solving that issue for us!!!

There appears to be a race condition that prevents udhcpd from running on
some occasions. Rebooting works around the issue.

I added a recent hack to lower the lease time dramatically to
work-around the issue.

I don’t think it’s the same issue. In my case, there’s no udhcpd process,
while systemd wrongly thinks that it’s running. I blame systemd, but
then, I always do :wink:

I’d be interested to hear what is the issue you’re having, as I might be
able to help (I’ve implemented DHCPv4 before).

I haven’t really narrowed it down, but I think on unclean shutdowns, and perhaps other conditions as well, the leases file doesn’t get deleted and UDHCPD thinks all (the only) lease is already checked out. It could also have to do with the virtual MAC address changing at times. Anyway, reducing the lease time is appearing to make the situation livable, but it isn’t a fix to the issue.

I think our lives could be a lot easier if we could figure out how to configure dnsmasq for all of our DNS/DHCP needs (and keep connman out of the way).

For newbs, wifi network config via a GUI is a big deal, but we are typically headless.

2

I think Robert would have to reply, but I'm guessing that connman provides a
GUI and lots of seemingly useful defaults while remaining relatively small.

Yes. See the note at the bottom.

Can /etc/network handle it all? How do you enable a GUI or even CUI to edit
it?

It cannot and you don't. See the note at the bottom.

I haven't really narrowed it down, but I think on unclean shutdowns, and
perhaps other conditions as well, the leases file doesn't get deleted and
UDHCPD thinks all (the only) lease is already checked out.

Right now my board is broken, so I cannot do more debugging -- as soon as
I do a wifi scan, it Oopses. I've reflashed it over USB, no help, so
I guess either my uBoot is corrupted, or something is wrong with my power
supply. I'll keep hacking (next step is to reflash from SD, advice
appreciated), but I'm soon running out of time.

But what I was seeing with udhcpcd is a different issue -- udhcpd does
a DHCPOFFER (the first "discovery" stage of lease acquisition), then gets
killed by SIGTERM. I'll go off a limb and blame systemd again, but
I really don't know.

I think our lives could be a lot easier if we could figure out how to
configure dnsmasq for all of our DNS/DHCP needs (and keep connman out of the
way).

Definitely. Dnsmasq is a rock solid piece of software.

For newbs, wifi network config via a GUI is a big deal,

I most respectfully disagree.

Newbies need clear instructions that work. Newbies are fine with editing
a cryptical file, as long as the result works reliably. What is not
user-friendly, on the other hand, is having a system that fails for
mysterious reasons -- newbies cannot troubleshoot it, so they try the same
operations again and again and tear their hair out.

Jason, I've got 20 years experience working with first year students.
They're 18, they're uneducated, and not all of them are CS students --
I work with biologists, psychology students, even linguistic students. And
the conclusion is always the same, clear and working instructions are
fine, no matter how cryptic, things that don't work reliably are
frustrating.

If I were you, I'd do something much simpler. Less ambitious, but rock
solid. I can outline it if you think it can be useful.

but we are typically headless.

Which only adds insult to injury.

-- Juliusz