Skip to contentSkip to conversation

nlif_index2name bug?

DISCUSSION Jaco Kroon 2 participants 4 messages Nov 17, 2007 – Nov 18, 2007

lore
Share & tools
lore

Jaco Kroon

author
#0 lore raw

Hi guys,

I've seen an error in nlif_index2name, specifically, if an interface comes up _after_ I've opened the nlif_handle (using nlif_open()) then it won't resolve the index of that device to a name.

Is this a known issue, a bug or simply me not understanding how nlif works? Currently I need to kill and restart my application in order to make it work again, which is obviously non-ideal.

Jaco

#1 lore raw

Hi,

Quoted 8 lines · Jaco Kroon

Hi guys,

I've seen an error in nlif_index2name, specifically, if an interface comes up _after_ I've opened the nlif_handle (using nlif_open()) then it won't resolve the index of that device to a name.

Is this a known issue, a bug or simply me not understanding how nlif works?

Yes, you did not understand how nlif is working (but I think this is due to the lack of documentation). If fact, you have omit to listen to iface events and to call nlif_catch after each event.

For a working code example, you can have a look at NuFW's code. nlif related code is always prefixed by:
#ifdef HAVE_NLIF_CATCH

You can browse code online at:
http://software.inl.fr/trac/trac.cgi/browser/mirror/edenwall/nufw/trunk/nufw/src/nufw/packetsrv.c
http://software.inl.fr/trac/trac.cgi/browser/mirror/edenwall/nufw/trunk/nufw/src/nufw/iface.c

I've just wrote a brief nlif documentation on the following page:
http://software.inl.fr/trac/trac.cgi/wiki/articles/using_nlif

BR,

Signature · 2 lines
Eric Leblond <[email protected]>
INL
author
#2 lore raw
Quoted 10 lines · Eric Leblond

Hi,

Quoted 6 lines · Jaco Kroon

Hi guys,

I've seen an error in nlif_index2name, specifically, if an interface comes up _after_ I've opened the nlif_handle (using nlif_open()) then it won't resolve the index of that device to a name.

Quoted 8 lines
Quoted 3 lines

Is this a known issue, a bug or simply me not understanding how nlif works?

Yes, you did not understand how nlif is working (but I think this is due to the lack of documentation). If fact, you have omit to listen to iface events and to call nlif_catch after each event.

Ah thanks.

Quoted 8 lines

For a working code example, you can have a look at NuFW's code. nlif related code is always prefixed by:
#ifdef HAVE_NLIF_CATCH

You can browse code online at:
http://software.inl.fr/trac/trac.cgi/browser/mirror/edenwall/nufw/trunk/nufw/src/nufw/packetsrv.c
http://software.inl.fr/trac/trac.cgi/browser/mirror/edenwall/nufw/trunk/nufw/src/nufw/iface.c

This isn't a particularly "short" example and can probably be simplified a lot, not true? It's good enough for me though, so thank a million for pointing me to it.

Quoted 3 lines

I've just wrote a brief nlif documentation on the following page:
http://software.inl.fr/trac/trac.cgi/wiki/articles/using_nlif

This explains things quite a bit. Very nicely done.

I'm actually using libnetfilter_queue, and am not doing anything multi-threaded, would it be sufficient just to call nlif_catch() every time I run into an interface for which there is no name and to then retry the call? Alternatively, you mention select(), I'm guessing the poll() system call will achieve the same thing? My reasoning here is that everytime just before calling nlif_index2name() to first call poll() with a timeout of zero in order to check whether there are updates and to then call nlif_catch if there are updates, however, is this the most efficient way, since poll() and nlif_catch() both probably results in a system call, wouldn't it simply be more efficient to just call nlif_catch every time just before calling nlif_index2name?

Also, in the case of poll(), I guess I need to poll for POLLIN?

Thanks for the help and explanation so far,

Jaco

author
#3 lore raw
Quoted 20 lines · Jaco Kroon
Quoted 4 lines · Eric Leblond

I've just wrote a brief nlif documentation on the following page:
http://software.inl.fr/trac/trac.cgi/wiki/articles/using_nlif

This explains things quite a bit. Very nicely done.

I'm actually using libnetfilter_queue, and am not doing anything multi-threaded, would it be sufficient just to call nlif_catch() every time I run into an interface for which there is no name and to then retry the call? Alternatively, you mention select(), I'm guessing the poll() system call will achieve the same thing? My reasoning here is that everytime just before calling nlif_index2name() to first call poll() with a timeout of zero in order to check whether there are updates and to then call nlif_catch if there are updates, however, is this the most efficient way, since poll() and nlif_catch() both probably results in a system call, wouldn't it simply be more efficient to just call nlif_catch every time just before calling nlif_index2name?

Quoted 2 lines

Also, in the case of poll(), I guess I need to poll for POLLIN?

Ok, so after testing it turns out poll(2) does work quite nicely, however, I need to call nlif_catch quite a few times to get things to work nicely, so now I've basically got this piece of code just before nlif_index2name():

while (poll(&nlif_poll, 1, 0) > 0) {
    log (LOG_INFO, "Calling nlif_catch due to device changes.");
    nlif_catch(nlif);
}

And it turns out that simply restarting ppp0 causes quite a few events:

[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Calling nlif_catch due to device changes.
[INFO]: Setting device ppp0(9) to cap 2.

Not sure what's causing all those events but it works correctly now, thanks.

Jaco