Miscellaneous sysopt Commands

sysopt commands are used to turn on some system features that are used for some very specific network behaviors. These commands allow the PIX to modify its behavior to respond to a unique set of circumstances, such as issues on the network with the path MTU or devices that do not follow the normal protocol sequences.

Examples of some of the most commonly used sysopt commands are provided in the following sections.

sysopt connection timewait Command

The sysopt connection timewait command is necessary to cater to end host applications whose default TCP terminating sequence is a simultaneous close instead of the normal shutdown sequence (see RFC 793). In a simultaneous close, both ends of the transaction initiate the closing sequence, as opposed to the normal sequence in which one end closes and the other end acknowledges before initiating its own closing sequence.

The default behavior of the PIX Firewall is to track the normal shutdown sequence and release the connection after two FINs and the acknowledgment of the last FIN segment. This quick release heuristic lets the PIX Firewall sustain a high connection rate.

When simultaneous close occurs two FINs are sent across the PIX, one from each end of the connection. In order to close the connections on both ends of the TCP connection using simultaneous close, both ends must receive acknowledgements of the FINs they have sent out. However, the PIX thinks that the closing sequence is a normal TCP close sequence rather than a simultaneous close sequence. So after seeing the two FINs and then the acknowledgement from one end, it closes out the connection on the firewall as it would if it were a normal close sequence. However, in this case, one end of the TCP connection is left waiting, in the CLOSING state, for a response to the FIN it has sent out. Many sockets in the CLOSING state can degrade an end host's performance. For instance, some WinSock mainframe clients are known to exhibit this behavior and degrade the performance of the mainframe server. Old versions of HP/UX are also susceptible to this behavior. Enabling the sysopt connection timewait command creates a quiet-time window for the abnormal close-down sequence to complete. What this means is that after the PIX has seen what is considers to be the close of a connection, it waits 15 seconds before it closes out the connections. This wait time allows simultaneously closing connections to close on both ends properly since hopefully the final ACK gets transmitted during the 15 seconds of grace period. Please note however, that the PIX waits 15 seconds for all connection closes once this feature has been turned on, irrespective of whether they are normal closes or simultaneous closes. This can result in significant degradation of performance in environments where there are a lot of new connections being created and a lot of old connections being terminated in large quantities. This is the reason this feature is turned off by default.

sysopt noproxyarp Command

The sysopt noproxyarp command allows you to disable proxy ARPs on a PIX Firewall interface.

Proxy ARP in PIX is used mainly when the PIX wants to receive packets destined for a certain IP address on one of its interfaces and forward them to another interface connected to it. An example of this is the PIX proxy ARPing for the addresses contained in PIX's global pool of NAT addresses for the hosts sitting on the inside private network behind the PIX, provided that an xlate has been created by traffic originating from the inside network. Upon receiving the packets destined for these addresses, the PIX forwards them to the inside network after performing the normal checks and translations.

In general, while Proxy ARPing a perfectly legitimate feature of networking, Proxy ARP should be best thought of as a temporary transition mechanism. Its use should not be encouraged as part of a stable solution. A number of potential problems are associated with its use, including hosts' inability to fall back to alternative routes if a network component fails, and the possibility of race conditions and bizarre traffic patterns if the bridged and routed network segments are not clearly delineated.

Using specific routes and default gateway routes eliminates the need for end hosts to use proxy ARPing as a mechanism for routing packets.

The sysopt noproxyarp command is generally used in situations where it is inappropriate for the PIX to respond with proxy ARPing due to network misconfigurations or bugs. An example of such a situation is shown in Figure 8-9.

Figure 8-9 sysopt nonproxyarp Feature Usage

Figure 8-9 sysopt nonproxyarp Feature Usage

Host A

The PIX in this scenario is set up with a nat (inside) 0 0 0 command, which implies that PIX will not translate any of the IP addresses on the inside network. Due to how the nat 0 command is designed, the PIX starts to proxy ARP on the outside interface for any and all addresses upon being configured in this manner. Consequently, if host A wants to access the Internet and sends an ARP and the PIX proxy ARPs before the router can respond, host A would be sending traffic destined for the Internet to the PIX, which is incorrect routing. The way around this scenario is to turn off proxy ARP on the PIX. However, after you do this, the PIX stops ARPing for all addresses, even the ones for which it should ARP. Now the router needs to be set up with routes for the inside network behind the PIX so that it does not have to rely on the PIX proxy ARPing for that network. This situation would never have arisen if the nat 0 command were set up with specific IP address ranges contained on the inside network. The noproxyarp command allows the system administrator to get over this mix of misconfiguration and product shortcoming.

sysopt connection tcpmss Command

The sysopt connection tcpmss command forces proxy TCP connections to have a maximum segment size no greater than a configurable number of bytes. This command requests that each side of a TCP connection not send a packet of a size greater than x bytes. The negotiation for mss is done during the initial TCP connection establishment.

The number of bytes can be a minimum of 28 and any maximum number. You can disable this feature by setting this value to 0. By default, the PIX Firewall sets 1380 bytes as the sysopt connection tcpmss even though this command does not appear in the default configuration. The calculation for setting the TCP maximum segment size to 1380 bytes is as follows:

1380 data + 20 TCP + 20 IP + 24 AH + 24 ESP_CIPHER + 12 ESP_AUTH + 20 IP = 1500 bytes

1500 bytes is the MTU for Ethernet connections. It is recommended that the default value of 1380 bytes be used for Ethernet and mixed Ethernet and Token Ring environments. If the PIX Firewall has all Token Ring interfaces, the MTU can be set to 4056. However, if even one link along the path through the network is not a Token Ring, setting the number of bytes to such a high value might cause poor throughput. In its 1380-byte default value, this command increases throughput of the sysopt security fragguard command.

Because the TCP maximum segment size is the maximum size that an end host can inject into the network at one time (see RFC 793 for more information on the TCP protocol), the tcpmss command allows for improved performance in network environments where the path MTU discovery is not taking place properly because an end host is not responding to requests to lower its MTU or a firewall is sitting next to the end host dropping the ICMP packets being sent to the end host to request a reduction in its MTU. The tcpmss command forces the size of the TCP segments to a small value during TCP's initialization sequence, thereby eliminating the need for ICMP type 3 code 4 messages to be used to try and reduce the MTU. Also, most end hosts, even the ones that do not respond to ICMP messages requesting a decrease in packet size, do respond to tcpmss negotiation favorably.

Was this article helpful?

+1 0

Post a comment