Security group connection tracking

Your security groups use connection tracking to track information about traffic to and from the instance. Rules are applied based on the connection state of the traffic to determine if the traffic is allowed or denied. With this approach, security groups are stateful. This means that responses to inbound traffic are allowed to flow out of the instance regardless of outbound security group rules, and vice versa.

As an example, suppose that you initiate a command such as netcat or similar to your instances from your home computer, and your inbound security group rules allow ICMP traffic. Information about the connection (including the port information) is tracked. Response traffic from the instance for the command is not tracked as a new request, but rather as an established connection, and is allowed to flow out of the instance, even if your outbound security group rules restrict outbound ICMP traffic.

For protocols other than TCP, UDP, or ICMP, only the IP address and protocol number is tracked. If your instance sends traffic to another host, and the host sends the same type of traffic to your instance within 600 seconds, the security group for your instance accepts it regardless of inbound security group rules. The security group accepts it because it’s regarded as response traffic for the original traffic.

When you change a security group rule, its tracked connections are not immediately interrupted. The security group continues to allow packets until existing connections time out. To ensure that traffic is immediately interrupted, or that all traffic is subject to firewall rules regardless of the tracking state, you can use a network ACL for your subnet. Network ACLs are stateless and therefore do not automatically allow response traffic. Adding a network ACL that blocks traffic in either direction breaks existing connections. For more information, see Network ACLs in the Amazon VPC User Guide.

Note

Security groups have no effect on DNS traffic to or from the Route 53 Resolver, sometimes referred to as the 'VPC+2 IP address' (see What is Amazon Route 53 Resolver? in the Amazon Route 53 Developer Guide), or the ‘AmazonProvidedDNS’ (see Work with DHCP option sets in the Amazon Virtual Private Cloud User Guide). If you wish to filter DNS requests through the Route 53 Resolver, you can enable Route 53 Resolver DNS Firewall (see Route 53 Resolver DNS Firewall in the Amazon Route 53 Developer Guide).

Untracked connections

Not all flows of traffic are tracked. If a security group rule permits TCP or UDP flows for all traffic (0.0.0.0/0 or ::/0) and there is a corresponding rule in the other direction that permits all response traffic (0.0.0.0/0 or ::/0) for any port (0-65535), then that flow of traffic is not tracked, unless it is part of an automatically tracked connection. The response traffic for an untracked flow is allowed based on the inbound or outbound rule that permits the response traffic, not based on tracking information.

An untracked flow of traffic is immediately interrupted if the rule that enables the flow is removed or modified. For example, if you have an open (0.0.0.0/0) outbound rule, and you remove a rule that allows all (0.0.0.0/0) inbound SSH (TCP port 22) traffic to the instance (or modify it such that the connection would no longer be permitted), your existing SSH connections to the instance are immediately dropped. The connection was not previously being tracked, so the change will break the connection. On the other hand, if you have a narrower inbound rule that initially allows an SSH connection (meaning that the connection was tracked), but change that rule to no longer allow new connections from the address of the current SSH client, the existing SSH connection is not interrupted because it is tracked.

Automatically tracked connections

Connections made through the following are automatically tracked, even if the security group configuration does not otherwise require tracking. These connections must be tracked to ensure symmetric routing, as there could be multiple valid reply paths.

• Egress-only internet gateways

• Gateway Load Balancers

• Global Accelerator accelerators

• NAT gateways

• Network Firewall firewall endpoints

• Network Load Balancers

• AWS PrivateLink (interface VPC endpoints)

• Transit gateway attachments

• AWS Lambda (Hyperplane elastic network interfaces)

Throttling

Amazon EC2 defines the maximum number of connections that can be tracked per instance. After the maximum is reached, any packets that are sent or received are dropped because a new connection cannot be established. When this happens, applications that send and receive packets cannot communicate properly. Use the conntrack_allowance_available network performance metric to determine the number of tracked connections still available for that instance type.

To determine whether packets were dropped because the network traffic for your instance exceeded the maximum number of connections that can be tracked, use the conntrack_allowance_exceeded network performance metric. For more information, see Monitor network performance for your EC2 instance.

With Elastic Load Balancing, if you exceed the maximum number of connections that can be tracked per instance, we recommend that you scale either the number of instances registered with the load balancer or the size of the instances registered with the load balancer.

Idle connection tracking timeout

The security group tracks each connection established to ensure that return packets are delivered as expected. There is a maximum number of connections that can be tracked per instance. Connections that remain idle can lead to connection tracking exhaustion and cause connections not to be tracked and packets to be dropped. You can set the timeout for idle connection tracking on an Elastic network interface.

Note

This feature is only available for instances built on the AWS Nitro System.

There are three configurable timeouts:

• TCP established timeout: Timeout (in seconds) for idle TCP connections in an established state. Min: 60 seconds. Max: 432000 seconds (5 days). Default: 432000 seconds. Recommended: Less than 432000 seconds.

• UDP timeout: Timeout (in seconds) for idle UDP flows that have seen traffic only in a single direction or a single request-response transaction. Min: 30 seconds. Max: 60 seconds. Default: 30 seconds.

• UDP stream timeout: Timeout (in seconds) for idle UDP flows classified as streams which have seen more than one request-response transaction. Min: 60 seconds. Max: 180 seconds (3 minutes). Default: 180 seconds.

You may want to modify the default timeouts for any of the following cases:

• If you are monitoring tracked connections using Amazon EC2 network performance metrics, the conntrack_allowance_exceeded and conntrack_allowance_available metrics enable you to monitor dropped packets and tracked connection utilization to proactively manage EC2 instance capacity with scale up or out actions to help meet network connections demand before dropping packets. If you are observing conntrack_allowance_exceeded drops on your EC2 instances, you may benefit from setting a lower TCP established timeout to account for stale TCP/UDP sessions resulting from improper clients or network middle boxes.

• Typically, load balancers or firewalls have TCP Established idle timeout in the range of 60 to 90 minutes. If you are running workloads that are expected to handle a very high number of connections (greater than 100k) from appliances like network firewalls, configuring a similar timeout on an EC2 network interface is advised.

• If you are running workloads with high numbers of connections like DNS, SIP, SNMP, Syslog, Radius and other services that primarily use UDP to serve requests, setting the ‘UDP-stream’ timeout to 60s provides higher scale/performance for existing capacity and to prevent gray failures.

• For TCP/UDP connections through network load balancers (NLBs) and elastic load balancers (ELB), all connections are tracked. Idle timeout value for TCP flows is 350secs and UDP flows is 120 secs, and varies from interface level timeout values. You may want to configure timeouts at the network interface level to allow for more flexibility for timeout than the defaults for ELB/NLB.

You have the option to configure the connection tracking timeouts when you do the following:

• Create a network interface

• Modify network interface attributes

• Launch an EC2 instance

• Create an EC2 instance launch template

Example

In the following example, the security group has inbound rules that allow TCP and ICMP traffic, and outbound rules that allow all outbound traffic.

Inbound
Protocol type	Port number	Source
TCP	22 (SSH)	203.0.113.1/32
TCP	80 (HTTP)	0.0.0.0/0
TCP	80 (HTTP)	::/0
ICMP	All	0.0.0.0/0

Outbound
Protocol type	Port number	Destination
All	All	0.0.0.0/0
All	All	::/0

With a direct network connection to the instance or network interface, the tracking behavior is as follows:

• Inbound and outbound TCP traffic on port 22 (SSH) is tracked, because the inbound rule allows traffic from 203.0.113.1/32 only, and not all IP addresses (0.0.0.0/0).

• Inbound and outbound TCP traffic on port 80 (HTTP) is not tracked, because the inbound and outbound rules allow traffic from all IP addresses.

• ICMP traffic is always tracked.

If you remove the outbound rule for IPv4 traffic, all inbound and outbound IPv4 traffic is tracked, including traffic on port 80 (HTTP). The same applies for IPv6 traffic if you remove the outbound rule for IPv6 traffic.