routing table broken. [not-quite-solved]

pappy_mcfae · Last edited by pappy_mcfae on Tue Apr 29, 2008 7:42 am; edited 2 times in total

The routing table on my old Toshiba has somehow broken itself. The only thing I have changed recently is an upgrade to baselayout-2/openrc. I avoided many of the pitfalls associated with this upgrade as I did it with the migration document open on another machine. The only thing I missed is re-writing my /etc/hosts file.

After the upgrade, I noticed that the network wasn't connecting properly. I finally got the wireless to work just in time to notice that my routing was all messed up.

Here is the result of route on the afflicted machine:

DarKRaveR · Posted: Thu Apr 24, 2008 10:10 am Post subject:

You can manipulate the routing table via 'route' or 'ip route'.

The routing table is usually in kernel, not a file. The two route outputs you posted are identical in their meaning - Where does the 127.0.0.1 show up?

I don't see anything being wrong in the routing table of the machine having problems.

krinn · Watchman Joined: 02 May 2003 Posts: 7471

pappy_mcfae · Posted: Thu Apr 24, 2008 6:48 pm Post subject:

On the good side of things, I did get the routing tables fixed, at least temporarily. The even better news is 2.6.25-gentoo-r1 supports ndiswrapper without whining. Oh happy day!

The bad news is my wireless setup remains screwed. I tried removing the parentheses, but it made no difference. The wireless network will not set itself for the operation I desire (static IP address 192.168.0.115, default routing through 192.168.0.1). It will give me an IP address of 192.168.0.102. In the process, it also rewrites /etc/resolv.conf. That means that somehow, dhcpcd is asserting itself and setting up a dynamic IP address; something I am directly telling it not to do.

Anyway, changing the /etc/conf.d/net file to

pappy_mcfae · Posted: Thu Apr 24, 2008 7:22 pm Post subject:

Um, I figured out what was wrong. It works if I tell it the right interface to use, to whit:

depontius · Advocate Joined: 05 May 2004 Posts: 3530

pappy_mcfae · Posted: Tue Apr 29, 2008 7:57 am Post subject:

I have decided that things aren't quite as solved as I hoped. When using ndiswrapper with 2.6.25-gentoo-r1 kernel, sometimes the kernel routing tables start improperly, as below:

Genewb · Apprentice Joined: 09 Jan 2007 Posts: 165

pappy_mcfae · Posted: Thu May 01, 2008 6:19 pm Post subject:

I was told about the move to baselayout2-openrc a few months before it hit. I was told at that time it might be a wise idea to go with iproute2. At the time, I did it more to be able to "fix" a bug involving the b43 wireless module in the .24 kernel family. Installing iproute2 was the only way I could get that interface to even come close to being cooperative. In the end, I abandoned that idea because the b43 was a complete POS, in my opinion. However, I kept iproute2 just because from the looks of it, that package had a lot more goodies and doo-dads to make working with the network a little more precise.

As to the problem described here, I did make a bug report. Here's hoping it goes somewhere.

Blessed be!
Pappy
_________________
This space left intentionally blank, except for these ASCII symbols.

Genewb · Apprentice Joined: 09 Jan 2007 Posts: 165

pappy_mcfae · Posted: Fri May 02, 2008 6:03 am Post subject:

At the time, I was upgrading to baselayout-2/openrc. I was also trying out the 2.6.25-gentoo-r1 kernel, AND trying to fix my automatic network script. My computers multi-task better than I, unfortunately.

The purpose of that script is to make sure that whether the computer boots with the wireless PCMCIA card, or the Tulip knock-off PCMCIA, the computer will automatically set up the proper network setup, and then set a static IP address to whichever interface comes up. Before baselayout-2, I could rename an interface in my /etc/conf.d/net file. After the upgrade, I still can rename the interface, but I have to rename it using local.start, and it has to be done after the interface is up, not before.

The problem I was having was that I had renamed wlan0 to eth0 in udev, but I was still invoking it as wlan0 in the wireless part of the network script. Long story short, the wireless interface wasn't being properly initialized. What I did to cure that was to change the interface name from wlan0 to eth0 in the script. Then, as if by magic, the wireless adapter worked properly...or so I thought.

Unfortunately as my tests today have shown, sometimes the 2.6.25-gentoo-r1 kernel starts the wireless adapter perfectly. It's ready to go as soon as I get done logging in. Other times, it takes somewhere around three minutes for the interface to come fully up. Other times, even when trying to cajole it into working, it simply will not come up. Talk about THE definition of intermittent!

So, distilling all that, the truth is I fixed one mistake, but I still have the bug in place. I'm still looking for a cure for the routing table problem, which is the thing that makes the interface work...or not. On that point, I'm still at a loss.

The good news is that once ndiswrapper starts with2.6.25-gentoo-r1, it is as rock solid as it is under the .22 family. The wireless interface on my test laptop is still up and running after ten and a half hours. That's worlds better than I could get out of the .23 or .24 kernel families...so at least we're heading in the right direction.

Hope that explains it.

Blessed be!
Pappy
_________________
This space left intentionally blank, except for these ASCII symbols.

soletan · n00b Joined: 08 Jan 2008 Posts: 4

depontius · Advocate Joined: 05 May 2004 Posts: 3530

Thanks for the example - I'd kind of gotten the impression that that was what I was going to need to do.

I think I'm going to wait until baselayout-2 actually goes stable before the upgrade. I may try one machine early - maybe.
_________________
.sigs waste space and bandwidth

Dagger · Retired Dev Joined: 11 Jun 2003 Posts: 765 Location: UK

pappy_mcfae it seems your network is _backgrounded_ and until the connection is set up your routing table gives you an error or network unreachable (because at that time there is NO connection). That also explains why sometimes your routing table works fine (connection came up fast - maybe even before all other services finished loading) and sometimes it takes minutes to set it up (it's retrying until it succeeds).
_________________
95% of all computer errors occur between chair and keyboard (TM)
Join the FSF as an Associate Member!
Post under CC license.

pappy_mcfae · Posted: Wed May 21, 2008 7:27 am Post subject:

Dagger · Retired Dev Joined: 11 Jun 2003 Posts: 765 Location: UK

1st of all... it has NOTHING to do with the kernel

It's all about your RC configuration. I would suggest asking how to disable wireless network backgrounding during the startup in your bug. Alternatively you ask send an email to roy (openrc creator) and he will be happy to provide you with all the answers you need (check my reply to your bug - you can find the roy's website and contact details there).

I'm pretty sure I already saw the answer to this question somewhere... just can't remember where.
_________________
95% of all computer errors occur between chair and keyboard (TM)
Join the FSF as an Associate Member!
Post under CC license.

pappy_mcfae · Posted: Wed May 21, 2008 6:35 pm Post subject: