Information About Connection Settings
This section describes why you might want to limit connections and includes the following topics:
TCP Intercept and Limiting Embryonic Connections
Limiting the number of embryonic connections protects you from a DoS attack. The ASA uses the per-client limits and the embryonic connection limit to trigger TCP Intercept, which protects inside systems from a DoS attack perpetrated by flooding an interface with TCP SYN packets. An embryonic connection is a connection request that has not finished the necessary handshake between source and destination. TCP Intercept uses the SYN cookies algorithm to prevent TCP SYN-flooding attacks. A SYN-flooding attack consists of a series of SYN packets usually originating from spoofed IP addresses. The constant flood of SYN packets keeps the server SYN queue full, which prevents it from servicing connection requests. When the embryonic connection threshold of a connection is crossed, the ASA acts as a proxy for the server and generates a SYN-ACK response to the client SYN request. When the ASA receives an ACK back from the client, it can then authenticate the client and allow the connection to the server.
Note When you use TCP SYN cookie protection to protect servers from SYN attacks, you must set the embryonic connection limit lower than the TCP SYN backlog queue on the server that you want to protect. Otherwise, valid clients can nolonger access the server during a SYN attack.
To view TCP Intercept statistics, including the top 10 servers under attack, see Chapter27, “Configuring Threat Detection”
Disabling TCP Intercept for Management Packets for Clientless SSL Compatibility
By default, TCP management connections have TCP Intercept always enabled. When TCP Intercept is enabled, it intercepts the 3-way TCP connection establishment handshake packets and thus deprives the ASA from processing the packets for clientless SSL. Clientless SSL requires the ability to process the 3-way handshake packets to provide selective ACK and other TCP options for clientless SSL connections. To disable TCP Intercept for management traffic, you can set the embryonic connection limit; only after the embryonic connection limit is reached is TCP Intercept enabled.
Dead Connection Detection (DCD)
DCD detects a dead connection and allows it to expire, without expiring connections that can still handle traffic. You configure DCD when you want idle, but valid connections to persist.
When you enable DCD, idle timeout behavior changes. With idle timeout, DCD probes are sent to each of the two end-hosts to determine the validity of the connection. If an end-host fails to respond after probes are sent at the configured intervals, the connection is freed, and reset values, if configured, are sent to each of the end-hosts. If both end-hosts respond that the connection is valid, the activity timeout is updated to the current time and the idle timeout is rescheduled accordingly.
Enabling DCD changes the behavior of idle-timeout handling in the TCP normalizer. DCD probing resets the idle timeout on the connections seen in the
show conn
command. To determine when a connection that has exceeded the configured timeout value in the timeout command but is kept alive due to DCD probing, the
show service-policy
command includes counters to show the amount of activity from DCD.
TCP Sequence Randomization
Each TCP connection has two ISNs: one generated by the client and one generated by the server. The ASA randomizes the ISN of the TCP SYN passing in both the inbound and outbound directions.
Randomizing the ISN of the protected host prevents an attacker from predecting the next ISN for a new connection and potentially hijacking the new session.
TCP initial sequence number randomization can be disabled if required. For example:
-
If another in-line firewall is also randomizing the initial sequence numbers, there is no need for both firewalls to be performing this action, even though this action does not affect the traffic.
-
If you use eBGP multi-hop through the ASA, and the eBGP peers are using MD5. Randomization breaks the MD5 checksum.
-
You use a WAAS device that requires the ASA not to randomize the sequence numbers of connections.
TCP Normalization
The TCP normalization feature identifies abnormal packets that the ASA can act on when they are detected; for example, the ASA can allow, drop, or clear the packets. TCP normalization helps protect the ASA from attacks. TCP normalization is always enabled, but you can customize how some features behave.
The TCP normalizer includes non-configurable actions and configurable actions. Typically, non-configurable actions that drop or clear connections apply to packets that are always bad. Configurable actions (as detailed in “Customizing the TCP Normalizer with a TCP Map” section) might need to be customized depending on your network needs.
See the following guidelines for TCP normalization:
-
The normalizer does not protect from SYN floods. The ASA includes SYN flood protection in other ways.
-
The normalizer always sees the SYN packet as the first packet in a flow unless the ASA is in loose mode due to failover.
TCP State Bypass
By default, all traffic that goes through the ASA is inspected using the Adaptive Security Algorithm and is either allowed through or dropped based on the security policy. The ASA maximizes the firewall performance by checking the state of each packet (is this a new connection or an established connection?) and assigning it to either the session management path (a new connection SYN packet), the fast path (an established connection), or the control plane path (advanced inspection). See the “Stateful Inspection Overview” section in the general operations configuration guide for more detailed information about the stateful firewall.
TCP packets that match existing connections in the fast path can pass through the ASA without rechecking every aspect of the security policy. This feature maximizes performance. However, the method of establishing the session in the fast path using the SYN packet, and the checks that occur in the fast path (such as TCP sequence number), can stand in the way of asymmetrical routing solutions: both the outbound and inbound flow of a connection must pass through the same ASA.
For example, a new connection goes to ASA 1. The SYN packet goes through the session management path, and an entry for the connection is added to the fast path table. If subsequent packets of this connection go through ASA 1, then the packets will match the entry in the fast path, and are passed through. But if subsequent packets go to ASA 2, where there was not a SYN packet that went through the session management path, then there is no entry in the fast path for the connection, and the packets are dropped. Figure 22-1 shows an asymmetric routing example where the outbound traffic goes through a different ASA than the inbound traffic:
Figure 22-1 Asymmetric Routing
If you have asymmetric routing configured on upstream routers, and traffic alternates between two ASAs, then you can configure TCP state bypass for specific traffic. TCP state bypass alters the way sessions are established in the fast path and disables the fast path checks. This feature treats TCP traffic much as it treats a UDP connection: when a non-SYN packet matching the specified networks enters the ASA, and there is not an fast path entry, then the packet goes through the session management path to establish the connection in the fast path. Once in the fast path, the traffic bypasses the fast path checks.
Guidelines and Limitations
Context Mode Guidelines
Supported in single and multiple context mode.
Firewall Mode Guidelines
Supported in routed and transparent mode.
Failover Guidelines
Failover is supported.
TCP State Bypass Unsupported Features
The following features are not supported when you use TCP state bypass:
-
Application inspection—Application inspection requires both inbound and outbound traffic to go through the same ASA, so application inspection is not supported with TCP state bypass.
-
AAA authenticated sessions—When a user authenticates with one ASA, traffic returning via the other ASA will be denied because the user did not authenticate with that ASA.
-
TCP Intercept, maximum embryonic connection limit, TCP sequence number randomization—The ASA does not keep track of the state of the connection, so these features are not applied.
-
TCP normalization—The TCP normalizer is disabled.
-
SSM and SSC functionality—You cannot use TCP state bypass and any application running on an SSM or SSC, such as IPS or CSC.
TCP State Bypass NAT Guidelines
Because the translation session is established separately for each ASA, be sure to configure static NAT on both ASAs for TCP state bypass traffic; if you use dynamic NAT, the address chosen for the session on ASA 1 will differ from the address chosen for the session on ASA 2.
Maximum Concurrent and Embryonic Connection Guidelines
Depending on the number of CPU cores on your ASA model, the maximum concurrent and embryonic connections may exceed the configured numbers due to the way each core manages connections. In the worst case scenario, the ASA allows up to
n
-1 extra connections and embryonic connections, where
n
is the number of cores. For example, if your model has 4 cores, if you configure 6 concurrent connections and 4 embryonic connections, you could have an additional 3 of each type. To determine the number of cores for your model, enter the
show cpu core
command.
Configuring Connection Settings
This section includes the following topics:
Customizing the TCP Normalizer with a TCP Map
To customize the TCP normalizer, first define the settings using a TCP map.
Detailed Steps
Step 1 To specify the TCP normalization criteria that you want to look for, create a TCP map by entering the following command:
ciscoasa(config)# tcp-map tcp-map-name
For each TCP map, you can customize one or more settings.
Step 2 (Optional) Configure the TCP map criteria by entering one or more of the following commands (see
Table 22-1
). If you want to customize some settings, then the defaults are used for any commands you do not enter.
Table 22-1 tcp-map Commands
|
|
check-retransmission
|
Prevents inconsistent TCP retransmissions.
|
checksum-verification
|
Verifies the checksum.
|
exceed-mss
{
allow
|
drop
}
|
Sets the action for packets whose data length exceeds the TCP maximum segment size.
(Default) The
allow
keyword allows packets whose data length exceeds the TCP maximum segment size.
The
drop
keyword drops packets whose data length exceeds the TCP maximum segment size.
|
invalid-ack
{
allow
|
drop
}
|
Sets the action for packets with an invalid ACK. You might see invalid ACKs in the following instances:
-
In the TCP connection SYN-ACK-received status, if the ACK number of a received TCP packet is not exactly same as the sequence number of the next TCP packet sending out, it is an invalid ACK.
-
Whenever the ACK number of a received TCP packet is greater than the sequence number of the next TCP packet sending out, it is an invalid ACK.
The
allow
keyword allows packets with an invalid ACK.
(Default) The
drop
keyword drops packets with an invalid ACK.
Note TCP packets with an invalid ACK are automatically allowed for WAAS connections.
|
queue-limit
pkt_num
[
timeout
seconds
]
|
Sets the maximum number of out-of-order packets that can be buffered and put in order for a TCP connection, between 1 and 250 packets. The default is 0, which means this setting is disabled and the default system queue limit is used depending on the type of traffic:
-
Connections for application inspection (the
inspect
command), IPS (the
ips
command), and TCP check-retransmission (the TCP map
check-retransmission
command) have a queue limit of 3 packets. If the ASA receives a TCP packet with a different window size, then the queue limit is dynamically changed to match the advertised setting.
-
For other TCP connections, out-of-order packets are passed through untouched.
If you set the
queue-limit
command to be 1 or above, then the number of out-of-order packets allowed for all TCP traffic matches this setting. For example, for application inspection, IPS, and TCP check-retransmission traffic, any advertised settings from TCP packets are ignored in favor of the
queue-limit
setting. For other TCP traffic, out-of-order packets are now buffered and put in order instead of passed through untouched.
The
timeout
seconds
argument sets the maximum amount of time that out-of-order packets can remain in the buffer, between 1 and 20 seconds; if they are not put in order and passed on within the timeout period, then they are dropped. The default is 4 seconds. You cannot change the timeout for any traffic if the
pkt_num
argument is set to 0; you need to set the
limit
to be 1 or above for the
timeout
keyword to take effect.
|
reserved-bits
{
allow
|
clear
|
drop
}
|
Sets the action for reserved bits in the TCP header.
(Default) The
allow
keyword allows packets with the reserved bits in the TCP header.
The
clear
keyword clears the reserved bits in the TCP header and allows the packet.
The
drop
keyword drops the packet with the reserved bits in the TCP header.
|
seq-past-window
{
allow
|
drop
}
|
Sets the action for packets that have past-window sequence numbers, namely the sequence number of a received TCP packet is greater than the right edge of the TCP receiving window.
The
allow
keyword allows packets that have past-window sequence numbers. This action is only allowed if the
queue-limit
command is set to 0 (disabled).
(Default) The
drop
keyword drops packets that have past-window sequence numbers.
|
synack-data
{
allow
|
drop
}
|
Sets the action for TCP SYNACK packets that contain data.
The
allow
keyword allows TCP SYNACK packets that contain data.
(Default) The
drop
keyword drops TCP SYNACK packets that contain data.
|
syn-data
{
allow
|
drop
}
|
Sets the action for SYN packets with data.
(Default) The
allow
keyword allows SYN packets with data.
The
drop
keyword drops SYN packets with data.
|
tcp-options
{
selective-ack
|
timestamp
|
window-scale
} {
allow
|
clear
}
Or
tcp-options range
lower upper
{
allow
|
clear
|
drop
}
|
Sets the action for packets with TCP options, including the selective-ack, timestamp, or window-scale TCP options.
(Default) The
allow
keyword
allows packets with the specified option.
(Default for
range
) The
clear
keyword clears the option and allows the packet.
The
drop
keyword drops the packet with the specified option.
The
selective-ack
keyword sets the action for the SACK option.
The
timestamp
keyword sets the action for the timestamp option. Clearing the timestamp option disables PAWS and RTT.
The
widow-scale
keyword sets the action for the window scale mechanism option.
The
range
keyword specifies a range of options. The
lower
argument sets the lower end of the range as 6, 7, or 9 through 255.
The
upper
argument sets the upper end of the range as 6, 7, or 9 through 255.
|
ttl-evasion-protection
|
Disables the TTL evasion protection. Do not enter this command it you want to prevent attacks that attempt to evade security policy.
For example, an attacker can send a packet that passes policy with a very short TTL. When the TTL goes to zero, a router between the ASA and the endpoint drops the packet. It is at this point that the attacker can send a malicious packet with a long TTL that appears to the ASA to be a retransmission and is passed. To the endpoint host, however, it is the first packet that has been received by the attacker. In this case, an attacker is able to succeed without security preventing the attack.
|
urgent-flag
{
allow
|
clear
}
|
Sets the action for packets with the URG flag. The URG flag is used to indicate that the packet contains information that is of higher priority than other data within the stream. The TCP RFC is vague about the exact interpretation of the URG flag, therefore end systems handle urgent offsets in different ways, which may make the end system vulnerable to attacks.
The
allow
keyword allows packets with the URG flag.
(Default) The
clear
keyword clears the URG flag and allows the packet.
|
window-variation
{
allow
|
drop
}
|
Sets the action for a connection that has changed its window size unexpectedly. The window size mechanism allows TCP to advertise a large window and to subsequently advertise a much smaller window without having accepted too much data. From the TCP specification, “shrinking the window” is strongly discouraged. When this condition is detected, the connection can be dropped.
(Default) The
allow
keyword allows connections with a window variation.
The
drop
keyword drops connections with a window variation.
|
Configuring Connection Settings
To set connection settings, perform the following steps.
Detailed Steps
|
|
|
Step 1
|
class-map
name
ciscoasa(config)# class-map bypass_traffic
|
Creates a class map to identify the traffic for which you want to disable stateful firewall inspection.
|
Step 2
|
match
parameter
ciscoasa(config-cmap)# match access-list bypass
|
Specifies the traffic in the class map. See the “Identifying Traffic (Layer 3/4 Class Maps)” section for more information.
|
Step 3
|
policy-map
name
ciscoasa(config)# policy-map tcp_bypass_policy
|
Adds or edits a policy map that sets the actions to take with the class map traffic.
|
Step 4
|
class
name
ciscoasa(config-pmap)# class bypass_traffic
|
Identifies the class map created in Step 1
|
Step 5
|
Do one or more of the following:
|
|
set connection
{[
conn-max
n
] [
embryonic-conn-max
n
] [
per-client-embryonic-max
n
] [
per-client-max
n
] [
random-sequence-number
{
enable
|
disable
}]}
ciscoasa(config-pmap-c)# set connection conn-max 256 random-sequence-number disable
|
Sets maximum connection limits or whether TCP sequence randomization is enabled.
The
conn-max
n
argument sets the maximum number of simultaneous TCP and/or UDP connections that are allowed, between 0 and 2000000. The default is 0, which allows unlimited connections.
If two servers are configured to allow simultaneous TCP and/or UDP connections, the connection limit is applied to each configured server separately.
When configured under a class, this argument restricts the maximum number of simultaneous connections that are allowed for the entire class. In this case, one attack host can consume all the connections and leave none of the rest of the hosts matched in the ACL under the class.
The
embryonic-conn-max
n
argument sets the maximum number of simultaneous embryonic connections allowed, between 0 and 2000000. The default is 0, which allows unlimited connections.
The
per-client-embryonic-max
n
argument sets the maximum number of simultaneous embryonic connections allowed per client, between 0 and 2000000. The default is 0, which allows unlimited connections.
The
per-client-max
n
argument sets the maximum number of simultaneous connections allowed per client, between 0 and 2000000. The default is 0, which allows unlimited connections. When configured under a class, this argument restricts the maximum number of simultaneous connections that are allowed for each host that is matched through an ACL under the class.
The
random-sequence-number
{
enable
|
disable
} keyword enables or disables TCP sequence number randomization. See the “TCP Sequence Randomization” section section for more information.
You can enter this command all on one line (in any order), or you can enter each attribute as a separate command. The ASA combines the command into one line in the running configuration.
Note For management traffic, you can only set the conn-max and embryonic-conn-max keywords.
|
|
set connection timeout
{[
embryonic
hh
:
mm
:
ss] {
idle
hh
:
mm
:
ss [
reset
]] [
half-closed
hh
:
mm
:
ss] [dcd hh
:
mm
:
ss [
max_retries
]]}
ciscoasa(config-pmap-c)# set connection timeout idle 2:0:0 embryonic 0:40:0 half-closed 0:20:0 dcd
|
Sets connection timeouts. For global timeouts, see the
timout
command in the command reference.
The
embryonic
hh
:
mm
:
ss
keyword
sets the timeout period until a TCP embryonic (half-open) connection is closed, between 0:0:5 and 1193:00:00. The default is 0:0:30. You can also set this value to 0, which means the connection never times out.
The
idle
hh
:
mm
:
ss
keyword sets the idle timeout period after which an established connection of any protocol closes, between 0:0:1 and 1193:0:0. The default is 1:0:0. You can also set this value to 0, which means the connection never times out. For TCP traffic, the
reset
keyword sends a reset to TCP endpoints when the connection times out.
The
half-closed
hh
:
mm
:
ss
keyword sets the idle timeout period until a half-closed connection is closed, between 0:5:0 (for 9.1(1) and earlier) or 0:0:30 (for 9.1(2) and later) and 1193:0:0. The default is 0:10:0. Half-closed connections are not affected by DCD. Also, the ASA does not send a reset when taking down half-closed connections.
The dcd keyword enables DCD. DCD detects a dead connection and allows it to expire, without expiring connections that can still handle traffic. You configure DCD when you want idle, but valid connections to persist. After a TCP connection times out, the ASA sends DCD probes to the end hosts to determine the validity of the connection. If one of the end hosts fails to respond after the maximum retries are exhausted, the ASA frees the connection. If both end hosts respond that the connection is valid, the ASA updates the activity timeout to the current time and reschedules the idle timeout accordingly. The retry-interval sets the time duration in
hh
:
mm
:
ss
format to wait after each unresponsive DCD probe before sending another probe, between 0:0:1 and 24:0:0. The default is 0:0:15. The max-retries sets the number of consecutive failed retries for DCD before declaring the connection as dead. The minimum value is 1 and the maximum value is 255. The default is 5.
The default
tcp
idle timeout is 1 hour.
The default
udp
idle timeout is 2 minutes.
The default
icmp
idle timeout is 2 seconds.
The default
esp
and
ha
idle timeout is 30 seconds.
For all other protocols, the default idle timeout is 2 minutes.
To never time out, enter 0:0:0.
You can enter this command all on one line (in any order), or you can enter each attribute as a separate command. The command is combined onto one line in the running configuration.
Note This command is not available for management traffic.
|
|
set connection advanced-options
tcp-map-name
ciscoasa(config-pmap-c)# set connection advanced-options tcp_map1
|
Customizes the TCP normalizer. See the “Customizing the TCP Normalizer with a TCP Map” section to create a TCP map.
|
|
set connection advanced-options tcp-state-bypass
ciscoasa(config-pmap-c)# set connection advanced-options tcp-state-bypass
|
Enables TCP state bypass.
|
Step 6
|
service-policy policymap_name {global | interface
interface_name
}
ciscoasa(config)# service-policy tcp_bypass_policy outside
|
Activates the policy map on one or more interfaces.
global
applies the policy map to all interfaces, and
interface
applies the policy to one interface. Only one global policy is allowed. You can override the global policy on an interface by applying a service policy to that interface. You can only apply one policy map to each interface.
|
Configuration Examples for Connection Settings
This section includes the following topics:
Configuration Examples for Connection Limits and Timeouts
The following example sets the connection limits and timeouts for all traffic:
ciscoasa(config)# class-map CONNS ciscoasa(config-cmap)# match any ciscoasa(config-cmap)# policy-map CONNS ciscoasa(config-pmap)# class CONNS ciscoasa(config-pmap-c)# set connection conn-max 1000 embryonic-conn-max 3000 ciscoasa(config-pmap-c)# set connection timeout idle 2:0:0 embryonic 0:40:0 half-closed 0:20:0 dcd ciscoasa(config-pmap-c)# service-policy CONNS interface outside
You can enter
set connection
commands with multiple parameters or you can enter each parameter as a separate command. The ASA combines the commands into one line in the running configuration. For example, if you entered the following two commands in class configuration mode:
ciscoasa(config-pmap-c)# set connection conn-max 600 ciscoasa(config-pmap-c)# set connection embryonic-conn-max 50
the output of the
show running-config policy-map
command would display the result of the two commands in a single, combined command:
set connection conn-max 600 embryonic-conn-max 50
Configuration Examples for TCP State Bypass
The following is a sample configuration for TCP state bypass:
ciscoasa(config)# access-list tcp_bypass extended permit tcp 10.1.1.0 255.255.255.224 any ciscoasa(config)# class-map tcp_bypass ciscoasa(config-cmap)# description "TCP traffic that bypasses stateful firewall" ciscoasa(config-cmap)# match access-list tcp_bypass ciscoasa(config-cmap)# policy-map tcp_bypass_policy ciscoasa(config-pmap)# class tcp_bypass ciscoasa(config-pmap-c)# set connection advanced-options tcp-state-bypass ciscoasa(config-pmap-c)# service-policy tcp_bypass_policy outside ciscoasa(config-pmap-c)# static (inside,outside) 209.165.200.224 10.1.1.0 netmask 255.255.255.224
Configuration Examples for TCP Normalization
For example, to allow urgent flag and urgent offset packets for all traffic sent to the range of TCP ports between the well known FTP data port and the Telnet port, enter the following commands:
ciscoasa(config)# tcp-map tmap ciscoasa(config-tcp-map)# urgent-flag allow ciscoasa(config-tcp-map)# class-map urg-class ciscoasa(config-cmap)# match port tcp range ftp-data telnet ciscoasa(config-cmap)# policy-map pmap ciscoasa(config-pmap)# class urg-class ciscoasa(config-pmap-c)# set connection advanced-options tmap ciscoasa(config-pmap-c)# service-policy pmap global