CentOS Stream 9

Option A: Azure portal

• Go to the Azure Marketplace and search for "CentOS Stream 9 on Azure SecureAnyCloud".
• Select the image and click Create.
• Configure the basics:
• Subscription: select your Azure subscription.
• Resource group: create a new one or select an existing one.
• Virtual machine name: enter a name for your VM.
• Region: select your preferred Azure region.
• Size: Standard_B2s is recommended.
• Under Administrator account, select SSH public key and enter your key.
• Under Inbound port rules, allow SSH (port 22).
• Click Review + create, then Create.

Option B: Azure CLI

Accept the Marketplace terms, then create the VM:

az vm image terms accept --urn sac1647283583153:centos-stream-9:default:latest
az vm create \
  --resource-group myResourceGroup \
  --name my-centos-stream-9-azure-vm \
  --image sac1647283583153:centos-stream-9:default:latest \
  --size Standard_B2s \
  --admin-username azureuser \
  --generate-ssh-keys \
  --public-ip-sku Standard

Once your VM is running, connect over SSH:

ssh azureuser@<public-ip-address>

To find your VM's public IP, run:

az vm show --resource-group myResourceGroup --name my-centos-stream-9-azure-vm --show-details --query publicIps -o tsv

After you connect, verify the operating system:

cat /etc/centos-release
cat /etc/os-release uname -r

Check disk space and memory:

df -h free -h

Verify that the Azure Linux Agent is running:

systemctl status waagent

CentOS Stream 9 is continuously delivered (rolling). Keeping it current pulls the latest package set and all security backports:

sudo dnf upgrade -y

The Azure network security group (NSG) is the primary inbound and outbound traffic control, and it is recommended for most deployments.

This minimal image does not ship a host firewall daemon (firewalld is not preinstalled), but the base iptables command is present. If you want a host firewall in addition to the NSG, install and enable firewalld first:

sudo dnf install -y firewalld
sudo systemctl enable --now firewalld

With firewalld running, inspect the active ruleset:

sudo firewall-cmd --list-all

To allow additional ports, run:

sudo firewall-cmd --permanent --add-port=80/tcp sudo firewall-cmd --permanent --add-port=443/tcp sudo firewall-cmd --reload

Option A: AWS Marketplace console

• Open the SecureAnyCloud CentOS Stream listing on AWS Marketplace and choose View purchase options, then Subscribe. One subscription covers every version in the listing.
• Once the subscription is active, choose Launch your software.
• Choose One-click launch from AWS Marketplace (or Launch from EC2 Console for full control), then in Version choose CentOS Stream 9 - YYYY-MM-DD (the date is when the version was added) and your Region.
• Instance type: m5.large is recommended; any current x86_64 type works. The software charge is $0.04 per vCPU per hour.
• Key pair: choose your EC2 key pair. It is the only way in. RSA and ED25519 keys both work.
• Security group: allow inbound TCP 22 from your own IP address or network only.
• Storage: the root volume is 10 GiB gp3 by default. Choose a larger size if you need one; cloud-init grows the root partition and filesystem to fill it on first boot.
• Metadata: leave the defaults. The image already requires IMDSv2. Then choose Launch.

Option B: AWS CLI

Run this on your own workstation. Replace the AMI ID with the one shown for your Region on the Marketplace launch page, and the key pair, security group and subnet with your own. The root device of this AMI is /dev/sda1:

aws ec2 run-instances \
  --image-id ami-0123456789abcdef0 \
  --instance-type m5.large \
  --key-name my-key-pair \
  --security-group-ids sg-0123456789abcdef0 \
  --subnet-id subnet-0123456789abcdef0 \
  --block-device-mappings 'DeviceName=/dev/sda1,Ebs={VolumeSize=20,VolumeType=gp3}' \
  --tag-specifications 'ResourceType=instance,Tags=[{Key=Name,Value=centos-stream-9}]'

You do not need --metadata-options: the AMI already makes IMDSv2 required.

From your workstation, connect as ec2-user with the key pair you chose:

ssh -i ~/.ssh/my-key-pair.pem ec2-user@<public-ip>

There is no password to enter. If SSH asks for one, or says Too many authentication failures, see Troubleshooting below.

You can also connect without opening port 22 at all through AWS Systems Manager Session Manager, if the instance has an instance profile that allows Systems Manager. The agent is installed and running on this image.

cat /etc/centos-release grep -E '^(NAME|VERSION|ID|VERSION_ID|PLATFORM_ID|PRETTY_NAME)=' /etc/os-release uname -srm

Expected output:

CentOS Stream release 9 NAME="CentOS Stream" VERSION="9" ID="centos" VERSION_ID="9" PLATFORM_ID="platform:el9" PRETTY_NAME="CentOS Stream 9" Linux 5.14.0-758.el9.x86_64 x86_64

CentOS Stream has no point releases: the release is just 9, and the system moves forward through ordinary dnf updates. Your kernel version will be newer if updates have installed since this image was built.

This version keeps the CentOS Stream 9 EC2 package set whole. The packages that the Minimal version removes are installed here, and tuned is running:

rpm -qa | wc -l rpm -q tuned linux-firmware geolite2-city geolite2-country systemctl is-active tuned tuned-adm active

Expected output (the package versions change over time):

516 tuned-2.28.0-1.el9.noarch linux-firmware-20260812-164.el9.noarch geolite2-city-20191217-6.el9.noarch geolite2-country-20191217-6.el9.noarch active Current active profile: virtual-guest

• tuned is the system tuning daemon. Its virtual-guest profile, the one tuned-adm recommend picks on EC2, sets memory writeback and swappiness, disk readahead and CPU performance settings for a guest on a hypervisor. To use another profile, list them with tuned-adm list (tuned also ships an aws profile) and switch with, for example, sudo tuned-adm profile throughput-performance.
• linux-firmware is the firmware collection for physical network, graphics and wireless hardware (about 700 MB installed). An EC2 instance does not load it, but it is part of the standard CentOS Stream package set, and tools that expect it find it.
• geolite2-city and geolite2-country are IP geolocation databases that some tools can use.
• The sos support-report collector is not part of the CentOS Stream 9 EC2 base image. If a support engineer asks for a report, install it with sudo dnf -y install sos and run sudo sos report.

What the Minimal version drops. CentOS Stream 9 Minimal removes linux-firmware, tuned, geolite2-city and geolite2-country, and lists the removals in /usr/share/doc/sac/minimal-manifest.txt on its own image. A new Minimal instance has 512 packages against 516 here, and about 700 MB less in use on its root filesystem. Everything else is the same in both versions: the CentOS Stream 9 EC2 base image is already small, so neither version has sound firmware, Cockpit or Plymouth installed, and both carry the same hardening, compliance tooling and AWS agents. The package count on your instance moves by a few as kernels and updates install.

The evaluation the build ran after its reboot ships on the image:

ls -lh /usr/share/doc/sac/ python3 -c "import json; s=json.load(open('/usr/share/doc/sac/cis-l1-summary.json')); print('profile: ', s['profileLabel'], '(SCAP Security Guide', s['ssgVersion'] + ')'); print('scored rules: ', s['scored'], ' pass:', s['pass'], ' fail:', s['fail'], ' rate:', str(s['ratePct']) + '%'); print('waived: ', s['waived'], ' not applicable:', s['notapplicable']); print('evaluated at: ', s['evaluatedAt'])"

Expected output:

total 39M -rw-r--r--. 1 root root 37M Oct 20 23:08 cis-l1-arf.xml -rw-r--r--. 1 root root 2.3M Oct 20 23:08 cis-l1-report.html -rw-r--r--. 1 root root 1.2K Oct 20 23:08 cis-l1-summary.json -rw-r--r--. 1 root root 1.7K Oct 20 23:08 cis-l1-tailoring.xml -rw-r--r--. 1 root root 2.3K Oct 20 23:08 cis-l1-waivers.txt -rw-r--r--. 1 root root 1.6K Oct 20 23:04 scap-security-guide-LICENSE.txt profile: CIS Level 1 Server (SCAP Security Guide 0.1.82) scored rules: 256 pass: 255 fail: 1 rate: 99.6% waived: 6 not applicable: 33 evaluated at: 2026-10-20T23:08:20Z

• cis-l1-report.html is the full OpenSCAP report, rule by rule. Copy it to your workstation and open it in a browser.
• cis-l1-arf.xml is the same evaluation in the machine-readable ARF format, for compliance tooling.
• cis-l1-tailoring.xml is the exact tailoring used, so you can re-run the same evaluation (Step 6).
• The pass rate is pass / (pass + fail + error + unknown). Rules that are not applicable to this machine, not selected, or waived are not counted.

To copy the report to your workstation, run this on the workstation:

scp -i ~/.ssh/my-key-pair.pem ec2-user@<public-ip>:/usr/share/doc/sac/cis-l1-report.html .

The six waived rules

grep -v '^#' /usr/share/doc/sac/cis-l1-waivers.txt | cut -f1,2

Expected output:

partition_for_tmp [single-volume-partitions] grub2_password [serial-console-and-boot] ensure_root_password_configured [serial-console-and-boot] accounts_maximum_age_login_defs [key-only-account-aging] account_disable_post_pw_expiration [key-only-account-aging] sshd_limit_user_access [ssh-access-list]

The full reason for each is the third column of the same file. In short:

• Separate /tmp partition: this version boots from one EBS root volume. (The LVM versions of this listing provide separate volumes for /tmp, /var, /var/log, /var/log/audit, /var/tmp and /home.)
• Boot loader and root passwords: a GRUB or root password would have to be baked into the image, giving every instance the same secret. Root stays locked with no password. Set your own if you need serial console access (see Step 11).
• Password ageing defaults: the accounts cloud-init creates are key-only. An aging limit would expire them and lock the SSH key out. Accounts that do have a password are still aged.
• SSH AllowUsers/AllowGroups: a list fixed at build time would refuse accounts you create later. Your security group restricts who can reach port 22.

The one rule that does not pass

python3 -c "import json; [print(r) for r in json.load(open('/usr/share/doc/sac/cis-l1-summary.json'))['failingRules']]"

Expected output:

ensure_journald_and_rsyslog_not_active_together

journald and rsyslog together. The benchmark wants one logging system, not both. CentOS Stream runs journald and rsyslog together by default: journald keeps the structured system journal (persistent, in /var/log/journal), and rsyslog writes the traditional text logs (/var/log/messages, /var/log/secure, /var/log/cron) that many log shippers and audit procedures read. The image keeps both so neither kind of tooling loses its logs. If you standardize on one, for example the journal only, run sudo systemctl disable --now rsyslog on your instance and re-run Step 6.

You can repeat the evaluation at any time with the same content and tailoring. It takes under a minute:

sudo oscap xccdf eval \
  --tailoring-file /usr/share/doc/sac/cis-l1-tailoring.xml \
  --profile xccdf_co.sac_profile_cis-l1_ec2_single \
  --results /tmp/cis-check-results.xml --report /tmp/cis-check-report.html \
  /usr/share/xml/scap/ssg/content/ssg-cs9-ds.xml > /tmp/cis-check.log 2>&1 echo "oscap exit code: $? (2 = at least one rule did not pass)"
sudo python3 -c "import re,collections; c=collections.Counter(re.findall(r'<result>(\w+)</result>', open('/tmp/cis-check-results.xml').read())); s=sum(c[k] for k in ('pass','fail','error','unknown')); print(dict(sorted(c.items()))); print('pass rate: %.1f%%' % (100.0*c['pass']/s))"

Expected output:

oscap exit code: 2 (2 = at least one rule did not pass) {'fail': 1, 'notapplicable': 33, 'notselected': 1251, 'pass': 255} pass rate: 99.6%

Your own report is at /tmp/cis-check-report.html. Use the ssg-cs9-ds.xml content shown above: the scap-security-guide package also ships ssg-rhel9-ds.xml, but that content only applies to Red Hat Enterprise Linux, so on CentOS Stream it reports almost every rule as not applicable. As you install software and change settings, re-run this to see how your changes affect the result.

curl -s -o /dev/null -w 'IMDSv1 request without a token: HTTP %{http_code}\n' http://169.254.169.254/latest/meta-data/instance-id
TOKEN=$(curl -s -X PUT -H 'X-aws-ec2-metadata-token-ttl-seconds: 60' http://169.254.169.254/latest/api/token)
curl -s -H "X-aws-ec2-metadata-token: $TOKEN" -w '\n' http://169.254.169.254/latest/meta-data/placement/region

Expected output (your Region may differ):

IMDSv1 request without a token: HTTP 401 us-east-1

A request without a session token is refused with HTTP 401, and the same request with a token works. Scripts and SDKs on the instance must use IMDSv2; every current AWS SDK and the AWS CLI do this automatically.

sudo firewall-cmd --state sudo firewall-cmd --list-all sudo ss -tlnp

Expected output (the process IDs may differ):

running public (active) target: default icmp-block-inversion: no interfaces: eth0 sources: services: dhcpv6-client ssh ports: protocols: forward: yes masquerade: no forward-ports: source-ports: icmp-blocks: rich rules: State Recv-Q Send-Q Local Address:Port Peer Address:PortProcess LISTEN 0 128 0.0.0.0:22 0.0.0.0:* users:(("sshd",pid=1109,fd=7)) LISTEN 0 128 [::]:22 [::]:* users:(("sshd",pid=1109,fd=8))

• Only sshd listens, on port 22. Nothing else on the image opens a TCP port.
• The network interface is eth0: the official CentOS Stream 9 image for EC2 boots with net.ifnames=0, so the first interface keeps the traditional name.
• The public zone allows ssh and dhcpv6-client (replies to the instance's IPv6 DHCP client, which is how an instance in an IPv6-enabled subnet gets its address). The CentOS default cockpit service is removed from the zone; Cockpit is not installed.
• The loopback interface sits in firewalld's trusted zone, with rules that drop traffic claiming a loopback source from anywhere else (a CIS requirement).

Opening a port for your own service takes two steps: allow it in the instance's security group, and allow it in firewalld. For example, for a web server on 443, run on the instance:

sudo firewall-cmd --permanent --add-service=https sudo firewall-cmd --reload

If a port is open in the security group but not in firewalld (or the other way round), the traffic will not arrive.

systemctl is-active chronyd amazon-ssm-agent chronyc -n sources

Expected output (the timings vary):

active active MS Name/IP address Stratum Poll Reach LastRx Last sample =============================================================================== ^* 169.254.169.123 3 4 377 1 +757ns[ +44us] +/- 482us

The ^* marks the source chrony is using: 169.254.169.123, the Amazon Time Sync Service. It is reachable from every instance without internet access, so time stays correct in a private subnet too.

The AWS Systems Manager agent is installed from Amazon's own package at its current release (3.3.5390.0 on this build). It is not in the CentOS Stream repositories, so dnf does not update it; Systems Manager can keep it current for you (the AWS-UpdateSSMAgent document, or the Auto update SSM Agent setting in Fleet Manager).

The image is fully patched at build time. Check what is pending now and how automatic updates are set up:

sudo dnf -q check-update --refresh echo "dnf check-update exit code: $? (0 = nothing to update, 100 = updates available)" grep -E '^(upgrade_type|apply_updates|reboot) =' /etc/dnf/automatic.conf systemctl is-enabled dnf-automatic.timer systemctl cat dnf-automatic.timer | grep -E '^(OnCalendar|RandomizedDelaySec)' uname -r

Expected output on an instance launched soon after the image was built (later launches list the updates released since, with exit code 100):

dnf check-update exit code: 0 (0 = nothing to update, 100 = updates available) upgrade_type = default apply_updates = yes reboot = never enabled OnCalendar=*-*-* 6:00 RandomizedDelaySec=60m 5.14.0-758.el9.x86_64

How dnf-automatic is configured on this image:

• upgrade_type = default installs every available update, not only security fixes. CentOS Stream's repositories carry no security advisory data, so a security-only setting would install nothing at all.
• apply_updates = yes installs the updates, not just downloads them.
• The timer runs daily at 06:00 (instance time, UTC by default) with a random delay of up to an hour.
• reboot = never: a new kernel or core library is installed but takes effect only after you reboot, at a time you choose.

Because CentOS Stream moves ahead of RHEL and updates arrive continuously, test updates on a non-production instance first if your workload is sensitive to package changes. To stop automatic updates, run sudo systemctl disable --now dnf-automatic.timer; to update everything now, run:

sudo dnf -y upgrade

Reboot if a new kernel was installed:

sudo systemctl reboot

getenforce
sudo passwd -S root
sudo sshd -T | grep -E '^(permitrootlogin|passwordauthentication|pubkeyauthentication|maxauthtries|clientaliveinterval|banner) ' cat /etc/issue.net

Expected output:

Enforcing root LK 2024-02-07 0 99999 7 -1 (Alternate authentication scheme in use.) maxauthtries 4 clientaliveinterval 300 permitrootlogin no pubkeyauthentication yes passwordauthentication no banner /etc/issue.net Authorized users only. All activity may be monitored and reported.

• Enforcing means SELinux is on and enforcing the targeted policy. Keep it on; if your own software is denied, check sudo ausearch -m avc -ts recent rather than disabling SELinux.
• LK in the second column means the root account is locked.
• sshd -T prints the settings the SSH server actually uses, after every file in /etc/ssh/sshd_config.d/ has been read. maxauthtries 4 and clientaliveinterval 300 come from the CIS profile: SSH allows four authentication attempts per connection and checks idle sessions every five minutes.
• The banner text is shown before login.

EC2 Serial Console and recovery. No account on this image has a password, so the serial console cannot log in until you set one. If you want serial console access, set a password for your own user on your running instance (the password is yours and is not on the image):

sudo passwd ec2-user

df -hT / lsblk -o NAME,SIZE,TYPE,FSTYPE,MOUNTPOINT
sudo grep -E 'cc_growpart.*(resized|NOCHANGE)' /var/log/cloud-init.log | tail -1 swapon --noheadings --show | wc -l sudo grep -h aide /etc/crontab

Expected output with the default 10 GiB root volume (on a larger volume, Size is larger and the cloud-init line says resized):

Filesystem Type Size Used Avail Use% Mounted on /dev/nvme0n1p4 xfs 8.8G 2.2G 6.6G 26% / NAME SIZE TYPE FSTYPE MOUNTPOINT nvme0n1 10G disk ├─nvme0n1p1 1M part ├─nvme0n1p2 200M part vfat /efi ├─nvme0n1p3 1G part xfs /boot └─nvme0n1p4 8.8G part xfs / 2026-10-20 23:30:06,542 - cc_growpart.py[DEBUG]: '/' NOCHANGE: no change necessary (/dev/nvme0n1, 4) 0 05 4 * * * root /usr/sbin/aide --check

• Disk layout. This is the standard (not LVM) layout of the official CentOS Stream 9 image: the root filesystem is XFS on partition 4 of the root volume, and /boot is a separate 1 GiB XFS partition. Partition 1 is the small BIOS boot partition the boot loader uses, and partition 2 is an EFI system partition the image carries but does not boot from (the AMI boots in legacy BIOS mode). If you need separate filesystems for /var, /var/log, /tmp or /home, or want to grow storage by adding volumes, choose the CentOS Stream 9 LVM version of this listing instead.
• Disk growth. cloud-init grows the root partition and filesystem to the volume size on first boot. To grow later, increase the EBS volume size in the console or with aws ec2 modify-volume, then run sudo growpart /dev/nvme0n1 4 && sudo xfs_growfs / (or reboot).
• Swap. The image has no swap. Add a swap file yourself if your workload needs one.
• AIDE. The file integrity database is built from the image's own files, and the CIS profile schedules aide --check daily at 04:05 from /etc/crontab.

The instance uses the standard CentOS Stream 9 repositories (BaseOS, AppStream and Extras):

sudo dnf -q list --available nginx

Expected output (the version changes over time):

Available Packages nginx.x86_64 2:1.20.1-32.el9 appstream

Install and start software in the usual way, for example sudo dnf -y install nginx then sudo systemctl enable --now nginx, and open its port in the security group and firewalld (Step 8). The CRB repository (build and development dependencies) is installed but disabled; enable it with sudo dnf config-manager --set-enabled crb if a package needs it. The AWS CLI is not on the image; install the current version from AWS (the AWS CLI install guide) if you need it.

To add another administrator, create the account, add it to the wheel group and give it an SSH key:

sudo useradd -m -G wheel alice
sudo install -d -m 700 -o alice -g alice /home/alice/.ssh echo "ssh-ed25519 AAAA... alice@laptop" | sudo tee /home/alice/.ssh/authorized_keys
sudo chown alice:alice /home/alice/.ssh/authorized_keys && sudo chmod 600 /home/alice/.ssh/authorized_keys
sudo restorecon -R /home/alice/.ssh

Because SSH is key-only, a new account needs its key in authorized_keys (and restorecon gives the file the SELinux label sshd expects). The new account has no password, so sudo cannot authenticate it until you give it one with sudo passwd alice. SSH stays key-only; the password is used by sudo. (Only ec2-user, created by cloud-init, has password-free sudo.)