Description of problem: ip tunnel can't be bound to another device Version-Release number of selected component (if applicable): kernel-2.6.23.8-63.fc8 with latest iproute-2.6.22 How reproducible: always Steps to Reproduce: 1. create tunnel # ip tunnel add tunneltest0 mode ipip remote 10.0.0.1 dev eth0 2. try to change the bounding device from eth0 to eth1 # ip tunnel change tunneltest0 dev eth1 3. show the result # ip tunnel show tunneltest0 # tunneltest0: ip/ip remote 10.0.0.1 local any dev eth0 ttl inherit Actual results: The tunnel bounding can't be change on the other device. Expected results: The tunnel bounding can be change on the other device.
Our short investigation showed that the SIOCCHGTUNNEL case in net/ipv6/sit.c:ipip6_tunnel_ioctl does not care about the p->link at all. And it's probably the same for ipip and gre.
I took a look at net/ipv4/ipip.c (this one is relevant for the provided testcase). Tomas, you're right that the handler for SIOCCHGTUNNEL ioctl ignores p.link completely. I have a patch to allow re-binding the tunnel to another device. It also recalculates the MTU of the tunnel from the known MTU of the newly bound device. I believe it's the right thing to do in this case. I'll send the patch to netdev and if it's accepted, I'll make similar changes to gre and sit. (Assigning the bug to myself.)
Created attachment 285761 [details] allow rebinding of ipip tunnels
The patch was sent to netdev (http://www.spinics.net/lists/netdev/msg49844.html). It was accepted with a small modification. I submitted similar patches for GRE and SIT (http://www.spinics.net/lists/netdev/msg49983.html). Dave Miller applied all three patches to his git tree. They will be in Linux 2.6.25.
Thanks, Michal. If I'm not mistaken, no further modification to iproute should be needed.
Tomas, correct, no change is needed in iproute.