Hi Axel,
sorry for the late response. I missed the email and found it today. I understand your concerns and agree with that. The reason for all this was to get the information on every gateway change. I need the node ip of the gateway node at the time when the gateway is selected or deselected. At moment I have modified the batmand to call a script in both cases to setup /etc/resolv.conf which is important to always have a valid dns server. the main publich dns servers often are not accessible from different provider networks.
/stephan
Hi,
I can understand your need and agree that your idea with the return value makes the hole thing very flexible. But i am not sure if it makes sense to call the policy routing script even more than once because system calls like that can be quite expensive. And another problem i see comes with the return values itself. Currently the policy-routing feature is not aware of the return value of the called script. One reason for this is that the policy-routing-script is essentially operating in a loop waiting to be feeded with new commands via a pipe. Therefore the script itself does not terminate after being feeded and does not return anything. Another problem is that the C-function execv() (which is currently used) does not really support return values except in case of an error. As you suggested, the function system() could be used instead, but the manpage suggests to not use this function with suid privileges.
Regarding the cost (in terms of processing-time) i did some small experiences on a netgear wgt634u and a linksys WRT with openwrt which showed that configuring a route via a bash script (using the ip command) or by using system() to execute the ip command is up to 50 times more expensive than doing it directly (using netlink sockets). In my test adding and removing 100 route entries using netlink sockets took about 200ms while using the ip command it takes about 6-10 seconds. Even just calling a script with system() which does nothing else than return 0 takes about 50ms per call.
Therefore I am not sure if using the policy routing script in a large network and with slow devices is a good idea at all. And if the threat must be blocked during the execution of the script to wait for the return value it would be even worse.
What about introducing the possibility to define that the routing information which is forwarded to the policy routing script is only informative and is still applied by the daemon itself?
ciao, axel
On Donnerstag 18 September 2008, Stephan Enderlein (Freifunk Dresden) wrote:
Hi Axel,
thanks for your comments. At moment I have no much time to spend for batman development. We got a son two month ago and I'm currently enjoy him much. But I have a good idea concerning the routing script. The problem is that I like batman-exp to setup all routes as defined by parameters, but also want batman-exp to call the script. Batman may call the script twice. One call before and one after setting the routes. All should depend on the return code of the first call of the script. If the script returns "0", then batman should not set routes and also there is no need to call the script a second time. If the first call of the script returns "1", then batman should set the routes as defined by parameters and after it should call the script a second time (like pre/post scripts).
This allows me to get the routing information without need to setup all routings per hand. Bye setting a environment variable you can distingnuish if the script is called as pre or post script. This leads to the next solution/patch:
You should also add information about the gateway selection/changes/deselection to this script. Together with the modification above you can let batmand set the routes and update your resolv.conf to find the correct router that knows how to resolve dns requests. At moment you have to get the dns ip from dhcp or you should enter this as a fix value. But a fix value for this is bad if you build a firmware with a simple user interface. Many people don't know how dns works and what ip they should enter. If you have different ISP some dns server are not accessible.
At moment I have added my own patch where I call (system("gateway_scirpt")) each time the gateway tunnel is created or deleted. this is working perfectly.
/Stephan
Hi
On Montag 15 September 2008, freifunk@ddmesh.de wrote:
Hi,
Just applied your latest patches as well. Thanks for looking over the code. Virgin eyes stumble easier over nasty stuff. :-)
When you find some problems in batman, can you also apply those patches to the batman-experimental branch? At moment it is running without problems for freifunk dresden. But if the network is growing perhaps some issues may cause problems.
Over the time a reasonable part of the code structure of bmx and batman has forked pretty much. Therefore I am not sure if it would be easy to simply apply existing batman patches to the bmx branch. But be sure, whenever I am getting aware of critical bugs identified in the batman code which also apply to the bmx code, I'll fix them too. But for many current and older series of patches its simply not necessary. For example looking at the main recent bug-fixes
Regarding the debug thread: I have removed the debug thread completely about 2 weeks ago (due to ongoing problems with this thing) and integrated its functionality into the main thread. I could not see any benefit of having this threaded except constant syncronization problems. Unfortunately I could not commit it yet because of unfinished testing. But I'll do it this week.
The gw-kernel module: There are no gw-tunnel module problems with bmx simply because there is no support for this feature. Most existing bmx-mesh networks I am aware of are using the one-way-tunnel mode. It does not implement the black hole detection but still allows you to dynamically change the preferred gw. Compared to the two-way-tunnel it has less overhead, avoids tunneling from the gw to the client node (no need to optimize something which does not exist) and allows internet access with only one level of network address translation.
Packet aggregation: Have been implemented and activated by default in bmx about a year ago and seem to work quite reliable since then.
Problem with timing issues have been solved individually
And very important. BMX has continued to rely on the concept of a rolling metric based on the number of received OGMs via the best path. Many patches in the batman-0.3 brach were due to changing this concept to a dedicated metric field carried with each OGM.
By the way, I think a number of bugs in bmx and batman have been identified due to your hints. Thanks for that. If you want to intensify your work on the code and want commit patches directly just let us know.
Bye Stephan
Dipl.Informatiker(FH) Stephan Enderlein Freifunk Dresden
B.A.T.M.A.N mailing list B.A.T.M.A.N@open-mesh.net https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n
B.A.T.M.A.N mailing list B.A.T.M.A.N@open-mesh.net https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n
Dipl.Informatiker(FH) Stephan Enderlein Freifunk Dresden
B.A.T.M.A.N mailing list B.A.T.M.A.N@open-mesh.net https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n
B.A.T.M.A.N mailing list B.A.T.M.A.N@open-mesh.net https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n
--------------------------------------- Dipl.Informatiker(FH) Stephan Enderlein Freifunk Dresden