From: Amin Azez Subject: Re: BUG: More ipt_recent queries Date: Tue, 21 Feb 2006 16:28:17 +0000 Message-ID: <43FB3FA1.7020808@ufomechanic.net> References: <43FAF692.8030804@ufomechanic.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: To: netfilter-devel@lists.netfilter.org In-Reply-To: <43FAF692.8030804@ufomechanic.net> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: netfilter-devel-bounces@lists.netfilter.org Errors-To: netfilter-devel-bounces@lists.netfilter.org List-Id: netfilter-devel.vger.kernel.org Amin Azez wrote: > I'm concerned about ipt_recent where it removes entries from the list. > > Surly the move-up-and-close-the-gap while loop will never enter because > time_info[time_loc].time has just been set to 0 so that this clause of > the while loop: > > time_info[(time_loc+1) % ip_list_tot].time < time_info[time_loc].time) > > will always be false. > > Fuller code segment: > > location = hash_table[hash_result]; > hash_table[r_list[location].hash_entry] = -1; > time_loc = r_list[location].time_pos; > time_info[time_loc].time = 0; > time_info[time_loc].position = location; > > while((time_info[(time_loc+1) % ip_list_tot].time < > time_info[time_loc].time) && ((time_loc+1) % ip_list_tot) != > curr_table->time_pos) { > time_temp = time_info[time_loc].time; > time_info[time_loc].time = time_info[(time_loc+1)%ip_list_tot].time; > time_info[(time_loc+1)%ip_list_tot].time = time_temp; > time_temp = time_info[time_loc].position; > time_info[time_loc].position = > time_info[(time_loc+1)%ip_list_tot].position; > time_info[(time_loc+1)%ip_list_tot].position = time_temp; > r_list[time_info[time_loc].position].time_pos = time_loc; > r_list[time_info[(time_loc+1)%ip_list_tot].position].time_pos = > (time_loc+1)%ip_list_tot; > time_loc = (time_loc+1) % ip_list_tot; > } > > > I think we should set time_info[time_loc].time = 0; at the end of the > while loop? This is actually solvable by changeing the while loop condition: while(((time_loc+1) % ip_list_tot) != curr_table->time_pos) { ... But appears also to have another problem which is that curr_table->time_pos should be decreased (mod ip_list_tot of course) to take into account that the last item in the newly shuffled time_info is now vacant; otherwise this test: for I in `seq 1 100` ; \ do echo ${S}1.1.1.$I > /proc/net/ipt_recent/LIMIT5 ; \ done ; \ echo -1.1.1.5 > /proc/net/ipt_recent/LIMIT5 ; \ echo 1.1.1.200 > /proc/net/ipt_recent/LIMIT5 Leaves 99 values instead of 100 because the slot from 1.1.1.5 was not re-used. Inserting this line after the end of the while loop: curr_table->time_pos=(curr_table->time_pos+ip_list_tot-1)%ip_list_tot; solves the problem and causes shuffled-to-the-end spare slots in time_info to be re-used first. A side effect is that output of /proc/net/ipt_recent/* doesn't come out in time order any more; if there have been mid-list deletes and then further adds. We could fix this by having ip_recent_get_info iterate indirectly using curr_table->time_info instead of curr_table->table. Comments? I guess I'll send one big patch when I get it all closed off. Sam