Tom Eastep 2003-2009 Thomas M. Eastep Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation; with no Invariant Sections, with no Front-Cover, and with no Back-Cover Texts. A copy of the license is included in the section entitled GNU Free Documentation License. This article applies to Shorewall 4.0 and later. If you are running a version of Shorewall earlier than Shorewall 4.0.0 then please see the documentation for that release.

Shorewall accounting rules are described in the file /etc/shorewall/accounting. By default, the accounting rules are placed in a chain called accounting and can thus be displayed using shorewall[-lite] show -x accounting. All traffic passing into, out of, or through the firewall traverses the accounting chain including traffic that will later be rejected by interface options such as tcpflags and maclist. The columns in the accounting file are described in shorewall-accounting (5) and shorewall6-accounting (5). In all columns except ACTION and CHAIN, the values -, any and all are treated as wild-cards. The accounting rules are evaluated in the Netfilter filter table. This is the same environment where the rules file rules are evaluated and in this environment, DNAT has already occurred in inbound packets and SNAT has not yet occurred on outbound packets. Accounting rules are not stateful -- each rule only handles traffic in one direction. For example, if eth0 is your Internet interface, and you have a web server in your DMZ connected to eth1, then to count HTTP traffic in both directions requires two rules: #ACTION CHAIN SOURCE DESTINATION PROTOCOL DEST SOURCE # PORT PORT DONE - eth0 eth1 tcp 80 DONE - eth1 eth0 tcp - 80 Associating a counter with a chain allows for nice reporting. For example: #ACTION CHAIN SOURCE DESTINATION PROTOCOL DEST SOURCE # PORT PORT web:COUNT - eth0 eth1 tcp 80 web:COUNT - eth1 eth0 tcp - 80 web:COUNT - eth0 eth1 tcp 443 web:COUNT - eth1 eth0 tcp - 443 DONE web Now shorewall show web (or shorewall-lite show web for Shorewall Lite users) will give you a breakdown of your web traffic: [root@gateway shorewall]# shorewall show web Shorewall-1.4.6-20030821 Chain web at gateway.shorewall.net - Wed Aug 20 09:48:56 PDT 2003 Counters reset Wed Aug 20 09:48:00 PDT 2003 Chain web (4 references) pkts bytes target prot opt in out source destination 11 1335 tcp -- eth0 eth1 0.0.0.0/0 0.0.0.0/0 tcp dpt:80 18 1962 tcp -- eth1 eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:80 0 0 tcp -- eth0 eth1 0.0.0.0/0 0.0.0.0/0 tcp dpt:443 0 0 tcp -- eth1 eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:443 29 3297 RETURN all -- * * 0.0.0.0/0 0.0.0.0/0 [root@gateway shorewall]# Here is a slightly different example: #ACTION CHAIN SOURCE DESTINATION PROTOCOL DEST SOURCE # PORT PORT web - eth0 eth1 tcp 80 web - eth1 eth0 tcp - 80 web - eth0 eth1 tcp 443 web - eth1 eth0 tcp - 443 COUNT web eth0 eth1 COUNT web eth1 eth0 Now shorewall show web (or shorewall-lite show web for Shorewall Lite users) simply gives you a breakdown by input and output: [root@gateway shorewall]# shorewall show accounting web Shorewall-1.4.6-20030821 Chains accounting web at gateway.shorewall.net - Wed Aug 20 10:27:21 PDT 2003 Counters reset Wed Aug 20 10:24:33 PDT 2003 Chain accounting (3 references) pkts bytes target prot opt in out source destination 8767 727K web tcp -- eth0 eth1 0.0.0.0/0 0.0.0.0/0 tcp dpt:80 0 0 web tcp -- eth0 eth1 0.0.0.0/0 0.0.0.0/0 tcp dpt:443 11506 13M web tcp -- eth1 eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:80 0 0 web tcp -- eth1 eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:443 Chain web (4 references) pkts bytes target prot opt in out source destination 8767 727K all -- eth0 eth1 0.0.0.0/0 0.0.0.0/0 11506 13M all -- eth1 eth0 0.0.0.0/0 0.0.0.0/0 [root@gateway shorewall]# Here's how the same example would be constructed on an HTTP server with only one interface (eth0). READ THE ABOVE CAREFULLY -- IT SAYS SERVER. If you want to account for web browsing, you have to reverse the rules below. #ACTION CHAIN SOURCE DESTINATION PROTOCOL DEST SOURCE # PORT PORT web - eth0 - tcp 80 web - - eth0 tcp - 80 web - eth0 - tcp 443 web - - eth0 tcp - 443 COUNT web eth0 COUNT web - eth0 Note that with only one interface, only the SOURCE (for input rules) or the DESTINATION (for output rules) is specified in each rule. Here's the output: [root@mail shorewall]# shorewall show accounting web Shorewall-1.4.7 Chains accounting web at mail.shorewall.net - Sun Oct 12 10:27:21 PDT 2003 Counters reset Sat Oct 11 08:12:57 PDT 2003 Chain accounting (3 references) pkts bytes target prot opt in out source destination 8767 727K web tcp -- eth0 * 0.0.0.0/0 0.0.0.0/0 tcp dpt:80 11506 13M web tcp -- * eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:80 0 0 web tcp -- eth0 * 0.0.0.0/0 0.0.0.0/0 tcp dpt:443 0 0 web tcp -- * eth0 0.0.0.0/0 0.0.0.0/0 tcp spt:443 Chain web (4 references) pkts bytes target prot opt in out source destination 8767 727K all -- eth0 * 0.0.0.0/0 0.0.0.0/0 11506 13M all -- * eth0 0.0.0.0/0 0.0.0.0/0 [root@mail shorewall]# For an example of integrating Shorewall Accounting with MRTG, see http://www.nightbrawler.com/code/shorewall-stats/.

The structure of the accounting rules changes slightly when there are bridges defined in the Shorewall configuration. Because of the restrictions imposed by Netfilter in kernel 2.6.21 and later, output accounting rules must be segregated from forwarding and input rules. To accomplish this separation, Shorewall-perl creates two accounting chains: accounting - for input and forwarded traffic. accountout - for output traffic. If the CHAIN column contains -, then: If the SOURCE column in a rule includes the name of the firewall zone (e.g., $FW), then the default chain to insert the rule into is accountout only. Otherwise, if the DEST in the rule is any or all or 0.0.0.0/0, then the rule is added to both accounting and accountout. Otherwise, the rule is added to accounting only.

Traditionally, the root of the Shorewall accounting rules has been the accounting chain. Having a single root chain has drawbacks: Many rules are traversed needlessly (they could not possibly match traffic). At any time, the Netfilter team could begin generating errors when loading those same rules (that has happened). MAC addresses may not be used in the accounting rules. The accounting chain cannot be optimized when OPTIMIZE_ACCOUNTING=Yes. The rules may be defined in any order so the rules compiler must post-process the ruleset to ensure that there are no loops and to alert the user to unreferenced chains. Beginning with Shorewall 4.4.18, the accounting structure can be created with three root chains: accountin: Rules that are valid in the INPUT chain (may not specify an output interface). accountout: Rules that are valid in the OUTPUT chain (may not specify an input interface or a MAC address). accounting: Other rules. The new structure is enabled by sectioning the accounting file in a manner similar to the rules file. The sections are INPUT, OUTPUT and FORWARD and must appear in that order (although any of them may be omitted). The first non-commentary record in the accounting file must be a section header when sectioning is used. Beginning with Shorewall 4.4.20, the ACCOUNTING_TABLE setting was added to shorewall.conf and shorewall6.conf. That setting determines the Netfilter table (filter or mangle) where the accounting rules are added. When ACCOUNTING_TABLE=mangle is specified, the available sections are PREROUTING, INPUT, OUTPUT, FORWARD and POSTROUTING. Section headers have the form: SECTION section-name When sections are enabled: You must jump to a user-defined accounting chain before you can add rules to that chain. This eliminates loops and unreferenced chains. You may not specify an output interface in the PREROUTING and INPUT sections. In the OUTPUT and POSTROUTING sections: You may not specify an input interface You may not jump to a chain defined in the INPUT or PREROUTING sections that specifies an input interface You may not specify a MAC address You may not jump to a chain defined in the INPUT or PREROUTING section that specifies a MAC address. The default value of the CHAIN column is: accountin in the INPUT section accounout in the OUTPUT section accountfwd in the FORWARD section accountpre in the PREROUTING section accountpost in the POSTROUTING section Traffic addressed to the firewall goes through the rules defined in the INPUT section. Traffic originating on the firewall goes through the rules defined in the OUTPUT section. Traffic being forwarded through the firewall goes through the rules from the FORWARD sections. Here is a sample sectioned file that used Per-IP Accounting. In this example, the dmz net corresponds to a vserver zone so lives on the firewall itself. #ACTION CHAIN SOURCE DESTINATION PROTO DEST SOURCE USER/ MARK IPSEC # PORT(S) PORT(S) GROUP SECTION INPUT ACCOUNT(fw-net,$FW_NET) - COM_IF ACCOUNT(dmz-net,$DMZ_NET) - COM_IF SECTION OUTPUT ACCOUNT(fw-net,$FW_NET) - - COM_IF ACCOUNT(dmz-net,$DMZ_NET) - - COM_IF SECTION FORWARD ACCOUNT(loc-net,$INT_NET) - COM_IF INT_IF ACCOUNT(loc-net,$INT_NET) - INT_IF COM_IF

Sergiusz Pawlowicz has written a nice article that shows how to integrate Shorewall Accounting with collectd to produce nice graphs of traffic activity. The article may be found at http://collectd.org/wiki/index.php/Plugin:IPTables.

Shorewall 4.4.17 added support for per-IP accounting using the ACCOUNT target. That target is only available when xtables-addons is installed. This support has been successfully tested with xtables-addons 1.32 on: Fedora 14 Debian Squeeze OpenSuSE 11.3 and xtables-addons Version 1.21 on: Debian Lenny Information about xtables-addons installation may be found at here. Per-IP accounting is configured in shorewall-accounting (5) (it is currently not supported in IPv6). In the ACTION column, enter: ACCOUNT(table,network) where table is the name of an accounting table (you choose the name). All rules specifying the same table will have their per-IP counters accumulated in that table. network is an IPv4 network in CIDR notation. The network can be as large as a /8 (class A). One nice feature of per-IP accounting is that the counters survive shorewall restart. This has a downside, however. If you change the network associated with an accounting table, then you must shorewall stop; shorewall start to have a successful restart (counters will be cleared). Example: Suppose your WAN interface is eth0 and your LAN interface is eth1 with network 172.20.1.0/24. To account for all traffic between the WAN and LAN interfaces: #ACTION CHAIN SOURCE DEST ... ACCOUNT(net-loc,172.20.1.0/24) - eth0 eth1 ACCOUNT(net-loc,172.20.1.0/24) - eth1 eth0 This will create a net-loc table for counting packets and bytes for traffic between the two interfaces. The table is dumped using the iptaccount utility (part of xtables-addons): iptaccount [-f] -l net-loc Example: gateway:~# iptaccount -l net-loc libxt_ACCOUNT_cl userspace accounting tool v1.3 Showing table: net-loc Run #0 - 3 items found IP: 172.20.1.105 SRC packets: 115 bytes: 131107 DST packets: 68 bytes: 20045 IP: 172.20.1.131 SRC packets: 47 bytes: 12729 DST packets: 38 bytes: 25304 IP: 172.20.1.145 SRC packets: 20747 bytes: 2779676 DST packets: 27050 bytes: 32286071 Finished. gateway:~# For each local IP address with non-zero counters, the packet and byte count for both incoming traffic (IP is DST) and outgoing traffic (IP is SRC) are listed. The -f option causes the table to be flushed (reset all counters to zero) after printing. For a command synopsis: iptaccount --help /sbin/shorewall also supports a show ipa command (from my own gateway just after I flushed the counters using iptaccount -f -l.: gateway:~# shorewall show ipa Shorewall 4.4.18-Beta1 per-IP Accounting at gateway - Thu Feb 10 13:28:37 PST 2011 Showing table: loc-net IP: 172.20.1.146 SRC packets: 9 bytes: 574 DST packets: 9 bytes: 770 Showing table: dmz-net IP: 70.90.191.124 SRC packets: 243 bytes: 23726 DST packets: 248 bytes: 39036 IP: 70.90.191.125 SRC packets: 73 bytes: 10640 DST packets: 73 bytes: 4846 Showing table: fw-net IP: 70.90.191.121 SRC packets: 0 bytes: 0 DST packets: 4 bytes: 243 IP: 70.90.191.122 SRC packets: 11 bytes: 1338 DST packets: 8 bytes: 5465 IP: 70.90.191.123 SRC packets: 42 bytes: 4604 DST packets: 44 bytes: 10662 gateway:~#

Beginning with the 3.3 kernels, Netfilter supports a form of accounting (nfacct) that is triggered by iptables rules but that survives purging and/or reloading the Netfilter ruleset. Shorewall support for this form of accounting was added in Shorewall 4.5.7. As of this writing (late July 2012), Fedora 17 has partial support for this feature but not all. It is necessary to download and build the following: libnetfilter_acct nfacct The following Fedora packages are also required: libnetlink and libnetlink-dev libmnl and libmnl-dev The tarballs are available from the Netfilter download sites. The nfacct utility can create, delete and display nfacct objects. These named objects consist of a packet and byte counter. Packets matching those netfilter rules that use the nfacct match cause the packet and byte count in the object named in the match to be incremented. To use nfaccnt with Shorewall, use the NFACCT target. See shorewall-accounting(5) for details. The shorewall show nfacct command is a thin wrapper around the nfacct list command.

Beginning with Shorewall 4.6.5, it is possible to preserve all ip[6]tables packet and byte counters over restarts and reboots through use of the -C option. This option is available in several commands. save Causes the packet and byte counters to be saved along with the chains and rules. restore Causes the packet and byte counters (if saved) to be restored along with the chains and rules. If your iptables ruleset depends on variables that are detected at run-time, either in your params file or by Shorewall-generated code, restore will use the values that were detected when the ruleset was saved, which may be different from the current values. start With Shorewall and Shorewall6, the -C option only has an effect if the -f option is also specified. If a previously-saved configuration is restored, then the packet and byte counters (if saved) will be restored along with the chains and rules. If your iptables ruleset depends on variables that are detected at run-time, either in your params file or by Shorewall-generated code, -C will use the values that were detected when the ruleset was saved, which may be different from the current values. restart If an existing compiled script is used (no recompilation required) and if that script generated the current running configuration, then the current netfilter configuration is reloaded as is so as to preserve the current packet and byte counters. If your iptables ruleset depends on variables that are detected at run-time, either in your params file or by Shorewall-generated code, -C will use the values that were detected when the ruleset was previously started, which may be different from the current values. If you wish to (approximately) preserve the counters over a possibly unexpected reboot, then: Create a cron job that periodically executes 'shorewall save -C'. Specify the -C and -f options in the STARTOPTIONS variable in either /etc/default/shorewall ( /etc/default/shorewall6, etc.) or /etc/sysconfig/shorewall (/etc/sysconfig/shorewall6, etc.), whichever is supported by your distribution. Note that not all distributions include these files so you may have to create the one(s) you need.