On every firewall I've run, a rule ends in permit or deny. So when I built my first security group in the lab, I looked for the action field so I could block one address. There isn't one.
What a security group can do
A security group is a list of things you allow. Anything you haven't allowed is dropped. You can't write a rule that says deny, so you can't carve an exception out of a wider allow.
It is also stateful. If you allow a request in, the reply goes back out without a rule of its own.
The rule, in Terraform
Here is the ingress rule from the lab. There is no argument for an action, because allow is the only option. The closest thing to a deny is leaving the rule out.
resource "aws_security_group" "web" {
name = "web"
vpc_id = aws_vpc.lab.id
}
resource "aws_vpc_security_group_ingress_rule" "https" {
security_group_id = aws_security_group.web.id
cidr_ipv4 = "0.0.0.0/0"
from_port = 443
to_port = 443
ip_protocol = "tcp"
}Where the deny lives
To block a specific address you go up a level, to the network ACL on the subnet. A NACL has both allow and deny, and it reads the way an access list does: numbered rules, lowest first, first match wins. In Terraform that is aws_network_acl_rule, and it does have a rule_action.
| Security group | Network ACL | |
|---|---|---|
| Applies to | A network interface | A subnet |
| Rules | Allow only | Allow and deny |
| State | Stateful | Stateless |
| Evaluation | All rules together | Number order, first match |
What I do now
In the lab, security groups say who may reach each resource, and each subnet has one NACL for the rare address I need to block. Before I call a security group done, I check three things:
- Each rule names a source, a port and a protocol.
- No rule is wider than the thing it protects needs.
- The description says why the rule exists.

