On Mar 5, 2019, at 8:04 AM, Daniel Finger <df@ewetel.de> wrote:
I need to replicate Access-Accept, Access-Reject and Accounting-Request and Accounting-Response tickets to another system for lawful interception.
That's... not really good. Sending *response* packets to another system is a pretty terrible idea. e.g. Accounting-Response packets are empty, and therefore don't contain any useful information.
Actually I need to add some other attributes, but this can all be done in unlang and is already working.
I wanted to use rlm_replicate for this, but it did not replicate the response.
I created a patch for rlm_replicate: - using the module in Post-Auth will replicate the response. - use the original request->packet->id that the response matches the request - when replicating an accounting-request, send a response right away (is there a better way without changing the current behaviour?)
Is it possible to have this functionality included?
I don't think so. For one, sending *response* packets somewhere else is very bad. For another, there are issues with the patch.
@@ -125,7 +134,7 @@ static int replicate_packet(UNUSED void }
packet->code = code; - packet->id = fr_rand() & 0xff; + packet->id = request->packet->id;
That changes the behaviour for *all* packets. This is wrong. The patch should use the same ID only when needed. e.g. only for response packets.
@@ -191,11 +200,18 @@ static int replicate_packet(UNUSED void */ RDEBUG("Replicating list '%s' to Realm '%s'", fr_int2str(pair_lists, list, "<INVALID>"), realm->name); - if (rad_send(packet, NULL, home->secret) < 0) { + if (rad_send(packet, request->packet, home->secret) < 0) {
This is similarly wrong. You've made this change for *all* packets. Not just for the response packets. The second argument is the original request packet, and is used only for response packets. Adding the second argument for request packets is not correct.
REDEBUG("Failed replicating packet: %s", fr_strerror()); rcode = RLM_MODULE_FAIL; goto done; } + if (code == PW_CODE_ACCOUNTING_REQUEST) { + rcode = replicate_packet(instance, request, PAIR_LIST_REPLY, PW_CODE_ACCOUNTING_RESPONSE); + if (rcode != RLM_MODULE_OK) { + goto done; + } + + }
The same goes here. You're *always* replicating an Accounting-Response packet. Any patch needs to be much more conservative. It's OK to add functionality. It's not OK to change existing functionality. Alan DeKok.