OK. This where the "problem" comes from: /* * If we are near a reset then add the next * limit, so that the user will not need to * login again */ if (data->reset_time && (res >= (data->reset_time - request->timestamp))) { res = data->reset_time - request->timestamp; res += check_vp->vp_integer; } (that's rlm_sqlcounter.c line about 710 in 2.0.5) Sqlcounter was designed for time counters. When your allowance (limit - time used) is greater than the time left in the period, time left in the period is added to the limit and sent as reply. This will create problems for data counters since limit values are much greater than those for time counters. As a workaround I would suggest that you comment this out if you are using data counters. Perhaps there is a case for adding another configuration item to sqlcounter that would determine what is counted time or data. Then for data counters this can be skipped or replaced with something like: if (data->reset_time && (trigger >= (data->reset_time - request->timestamp))) { res += check_vp->vp_integer; } where trigger would be set to a minute for hourly counter and hour for daily, weekly and monthly counters. I am sorry but I don't know how to write a patch. Ivan Kalik Kalik Informatika ISP Dana 24/10/2008, "liran tal" <liransgarage@gmail.com> piše:
Hey Ivan
2008/10/24 <tnt@kalik.net>
It (daily sqlcounter) does the same in 2.0.5:
rlm_sqlcounter: Authorized user jagoda, check_item=10000000, counter=2635 rlm_sqlcounter: Sent Reply-Item for user jagoda, Type=Session-Timeout, value=10027850
Returns value that is greater than the limit. I am using noreset sqlcounter and that one works fine.
Thanks for confirming this on a more up to date version. Alan, this smells like a bug (unless we missed something along the way), should I open up a bug ticket? And what would be the chances it can be backported to 1.1.7?
Thanks, Liran.