Security advisories and checks

Match VMStack's signed security advisories against your nodes' packages, and review the findings of the daily node and panel security checks.

Security › Advisories and checks brings together two things: VMStack's signed security advisories, matched against the kernel and package versions of your nodes, and the findings of security checks that run on every node and on the panel itself. Findings are open, fixed or dismissed, and high-severity ones alert the administrators.

Advisories#

VMStack publishes an advisory feed at https://vmstack.eu/security/advisories.json, with a signature beside it. Each advisory names a CVE, a severity and CVSS score, the packages and versions it affects, the fixed versions, and what to do.

  • The panel reads the feed daily and on Check now. The feed must be signed with VMStack's release key; a feed with another signature, an expired feed, or one older than the last feed read (a replay) is refused. The panel then keeps the last good feed and shows why the read failed.
  • Each node's agent reports the versions of the running kernel's package, QEMU, libvirt, OpenSSH, OpenSSL, sudo, glibc, dnsmasq and systemd. Versions are compared the way dpkg or rpm compares them, epochs included.
  • An advisory that affects a node becomes a finding on that node, with the installed and fixed versions and the advisory's severity and fix.
  • While an advisory of high or critical severity affects a node, a red banner is shown on every administrator page, with See what to do and Remind me tomorrow (hides it for 24 hours; a new advisory brings it back).

If the page says "No advisory feed yet", VMStack has not published a feed at that address yet. Nothing is wrong on your side.

On a panel without internet access, the feed can be turned off with this environment setting in the panel's .env (/opt/vmstack/shared/.env):

VMSTACK_ADVISORIES_CHECK=false

VMSTACK_ADVISORIES_URL points the panel at another copy of the feed (it must carry the same signature). The panel caches its configuration, so a change to .env takes effect after the configuration is cached again, which every panel update does.

Node checks#

The agent scans its node 2 minutes after it starts and every 6 hours, read-only:

Check Looks for
Ports Listening ports other than what a node is expected to run (sshd, the agent, dnsmasq on the bridge, QEMU console and migration ports, time and DHCP clients)
SSH Root password login, password sign-in, empty passwords
Updates Pending security updates, a pending reboot, stale package lists
Sudo Passwordless sudo rules
KVM /dev/kvm missing or open to everyone
Firewall No node firewall
Files World-writable files in the agent's directories
Modules Kernel modules loaded since the first scan that are not on the list of modules a KVM node normally loads

Pending security updates are medium severity; after they have waited 3 days they are raised to high, which alerts. The known-good module list keeps the modules check quiet: bridge, vhost, tap, netfilter and the like are expected on a KVM node. File contents (such as /etc/shadow) are not watched.

Node checks need agent 0.31 or later; older agents are listed as needing an update. See Agent updates.

Panel checks#

The panel checks itself, without calling any node:

Finding Severity
An administrator without a second factor (one finding per administrator) High
Debug mode on High
APP_URL not HTTPS High
The environment is not production Medium
No backup destination Medium
Two-factor authentication not required for administrators Low
An administrator's API key without allowed addresses (one per key) Low

Findings#

Each problem in each place is one finding, with its severity, detail, how to fix it, evidence, and when it was first and last seen.

State Meaning
Open Seen at the last check.
Fixed Not seen by the last check that ran. A check that could not run changes nothing. A fixed finding seen again opens again.
Dismissed An administrator dismissed it with a reason. It stays dismissed while it is still seen, until someone reopens it.

Filter the list by state, place (the panel or a node) and severity.

To dismiss a finding that is expected on your setup:

  1. Click Dismiss on the finding.
  2. Enter the reason, for example "port 8080 is our monitoring agent, filtered upstream". A reason is required.
  3. Confirm. The dismissal is audited (security.finding_dismissed).

To undo it, click Reopen.

When checks run#

  • Daily at 05:40 (vmstack:security-check): the panel checks, the feed is read, and each online node's latest report (up to 6 hours old) is read.
  • Check now on the page queues the same, asking each node for a report at most a minute old. The page updates as nodes answer, for 5 minutes.

Node reports are read by the nodes queue workers; if nothing updates, check the workers (Workers and queues).

Alerts#

A finding of high or critical severity that opens, reopens or is raised to high alerts the administrators: one grouped mail per burst, in the notification group Security findings of high or critical severity, sent at once by default. Fixing or dismissing it clears the alert. Raised findings are listed on the dashboard. See Alerts and notifications.

API#

Administrators can read findings and advisories with the scope security:read:

curl -s "https://panel.example.com/api/v1/security/findings?filter[state]=open&filter[severity]=high" \
  -H "Authorization: Bearer YOUR_API_KEY"

GET /api/v1/security/advisories returns the feed's status and every advisory with the nodes it affects.

Not supported#

  • Detecting live kernel patching (KernelCare and similar).
  • Listing the affected servers inside a node; advisories are matched per node.
  • Watching file contents, and malware scanning.
  • Per-node allowed ports (dismiss the finding instead).
  • Webhooks for findings.