Open-Mesh » batman-adv

Distribution: Fedora 24
Kernel: 4.7.9-200.fc24.x86_64

The crash with my environment is allways reproducible.

The Layout for the network devices is as follow:

I use a veth pair for communication with the device put to the network name space ( veth1 <-> eth0 )

Within the network name space I have

[br-wan]           comunication with main system
   [eth0]

[br-client]           
   [bat0]             
      [mesh-vpn]      tap device from fastd
      [mesh0]         usb wifi virtual device as MP
   [wlan0]            same usb device as AP            
   [eth1]             usb ethernet device

Debug fs is mounted for the name space.

Some applications are running withn the name space (fastd, alfred, alfred-vis, host_apd, thhtpd and a few others).

According to the informations one the monitot a page fault occur within sys_remove_file which is called from:

batadv_sysfs_remove_file_ns
batadv_softif_destroy vlan
batadv_softif_destroy_netlink

I don't use quemu, my implementation is very light.

There is no such thing as batadv_sysfs_remove_file_ns. Maybe you are using a patched version of batman-adv?

Can you please test with the current master (because it contains the missing patches for the netns support)? And can you please post the complete log.

And can you please post the relevant changes from my example to reproduce the problem? Please don't post the complete guide to setup this freifunk node because it most likely contains a lot of redundant things not relevant for this problem.

PS: qemu is only used by me so I can test around on a virtual system without destroying my development environment. It is not really relevant for netns because the netns stuff is done inside the qemu and doesn't know that it is inside a virtual environment.

PPS: the optimizer of gcc can turn the kernel backtraces to garbage. So it can be useful to compile the kernel/modules with a less aggressive optimization step to make them readable:

make EXTRA_CFLAGS="-fno-inline -O1 -fno-optimize-sibling-calls" KERNELPATH=/home/sven/tmp/linux-next V=1

Thought a little bit about what could cause this (assuming that you meant sysfs_remove_file_ns). And my best guess is that a bugfix for a leak is triggering a parallel removal of the bat0 device vlans. This conceptual problem was fixed using two commits in next:

• https://git.open-mesh.org/batman-adv.git/commit/a1f0a804cacda8ca0847cbac07adcbf95e9c24c1
• https://git.open-mesh.org/batman-adv.git/commit/0994697e926d9633adb4912036548b0bb1d110ea

So you can also first test the next branch (which is basically 2016.4 and doesn't contain the newest, not so well tested patches). But you should also know that 2016.4 is not providing debugfs in a non-init netns anymore. So you should also upgrade batctl to its current next branch (v2016.4) and alfred to its master branch (v2016.4).

I know that batman 2016.4 don't work with my actual code. I had downloaded it from kernel.org made a short test and reverted to 2016.3.
2016.4 seem to be promising bat I assume that we will have to wait a long time before ist is integrated within openwrt/led gluon.
How can I get the 2016 branch ? Git list only the master brancg.

There is no 2016 branch. And v2016.4 has to be tagged (maybe Simon tag the next branch as v2016.4 in the coming days).

And the time it takes until 2016.4 will be available in LEDE/OpenWrt depends on the loudness of our screams towards Marek. Not sure why he didn't accept my pull request for 2016.3 yet. I hope he will accept 2016.4 faster in openwrt-routing

And I will be waiting for gluon until Matthias finished its LEDE port. But you can get a first glance at it by using https://github.com/FreifunkVogtland/gluon/tree/v2016.1.x-hwtest. Freifunk Vogtland is already using it on their nodes (supernodes are still running 2016.3)

Btw. why did you download anything from kernel.org (there is no final Linux release which includes batman-adv 2016.4)? Our stuff can be found under https://git.open-mesh.org/ and https://downloads.open-mesh.org/ (v2016.4 is not yet available as tarball).

Easiest way to download it:

git clone -b next https://git.open-mesh.org/batman-adv.git
make -C batman-adv
sudo make -C batman-adv install
reboot
....
batctl -v

The same can be done with the batctl repo (not sure where you install it on your system) and the master branch of alfred.

And what didn't work? Did it still crash?

batctl -v tell 2016.3-41-gb7c5cd4

The kernel.org version is 2016.3-63-g061486b
(strings ./net/batman-adv/batman-adv.ko | grep '2016\.')

For both I have no more a crash.

by the way I suspect that there is an other problem, see:
https://forum.freifunk.net/t/router-advertisement-an-absender-zurueck/13714