Atrás 
Introduction
“The more things change, the more they stay the same.“
In the recent Equinix breach in September 2020, 74 RDP servers were exposed to the Internet. Any publicly exposed ports are a risk but remote access protocols such as RDP have had their share of critical vulnerabilities (e.g., BlueKeep in 2019).
In this blog, we will look at remote access statistics of public cloud workloads based on 287,877 compute instances across 327 anonymized production environments in AWS, Google Cloud, and Azure. The focus will be on a few common ports/protocols used for remote access or management of workloads, namely: SSH, RDP, and to a lesser extent, VNC.
What we will find is that:
- Direct Access is Still Very Common
Direct access to compute instances is still very common (35-85%+ of public workloads depending upon cloud provider environment)—allowing inbound traffic to ports from public CIDRs for SSH/RDP. Although this finding is not necessarily surprising, the high percentage of workloads is, from 35% up to 85%. Unsurprisingly, SSH is the most common due to the popularity of Linux workloads, followed by RDP, then VNC.
- Broad Internet Exposure is Alarmingly Common
Additionally, a fair number of network configurations allow broad source IP CIDRs to access these same ports (SSH/RDP) e.g. from the entire public Internet (0.0.0.0/0). Over 13% of AWS public instances allow inbound access to All Destination Ports from any public Internet address, 71% of AWS public instances allow SSH from any public Internet address, and 14% of AWS public instances allow RDP from any public Internet IP address.
- Better Secure Access Alternatives Not Deployed
The implication from the above is that better secure access alternatives from the cloud service providers or other vendors are not being deployed. These alternatives are more secure than direct access or bastion hosts in almost every area (credential/key management, authorization, auditing, protocol/port attack surface, protocol vulnerabilities) and are referenced later in this blog.
Direct remote access
To identify direct access, we looked at public compute instances (with at least one assigned public IP address) that have a network security group or firewall ruleset that allows inbound traffic to a port range that included any of: 22 (SSH), 3389 (RDP), or 3800/3900 (VNC) and from a public source IP range. We counted separately the All Port range (0-65535).
We might guess that remote access directly to public instances is still common in the cloud, but the frequency of occurrence is eye-opening:
- In AWS, out of 6,597 public EC2 instances, 16% allowed inbound traffic to All Ports, 36% to SSH, 8% to RDP, 1% to VNC.
- In GCP, out of 5,675 public compute instances, 55% allowed inbound traffic to All Ports, 88% to SSH, 85% to RDP, and 1% to VNC.
- In Azure, out of 15,432 public compute instances, 53% allowed inbound traffic to SSH.
As we can see, direct inbound access from the Internet to public compute instances is very common, regardless of cloud service provider. This reflects bad security practices since there are better alternatives mentioned below.
In addition, allowing traffic to all All Destination Ports (0-65535) is commonly found in AWS (16%) and GCP (55%) public compute instances. Overly broad port access only increases the risk of port scans and exploits of other services running on the compute instances that normally should not be exposed to the Internet.
Internet exposure
In the inbound traffic rules