Security pack
Hosting and operations
The server
- A dedicated virtual machine (Ubuntu Server LTS) on a server owned by IMS Creator, in Australia, used for nothing else that holds customer information.
- Its data disk is encrypted (LUKS2). Customer data, the temporary drawing area, secrets and logs are all on it.
- No port is open to the internet: the machine opens one outbound, encrypted tunnel to Cloudflare, and Cloudflare passes requests to it. The router has no port forwarded.
- It sits on its own network segment. Its firewall (nftables) lets SSH in only from the private administration network (Tailscale), with keys only, no passwords and no root sign-in. The containers cannot reach the owner's home or management networks: rules rebuilt each time Docker starts keep them out.
- The database and the file store listen only inside the machine; neither is published.
- Security updates are applied automatically (unattended upgrades) for the operating system. The application is updated from its own tested images, with rollback (see Application security).
At the edge (Cloudflare)
- TLS 1.2 or later, HSTS, and Cloudflare's managed firewall rules and bot protection in front of the application.
- Rate limits on the sign-in steps, on top of the application's own.
- A bot check (Cloudflare Turnstile) on the sign-in, password reset, registration and sign-up pages.
- Our administrators' own access goes through Cloudflare Access, as themselves.
Watching over it
- The application checks its own health (the database, the file store, free disk space) and the console shows it, with alerts for anything wrong: a process down, a disk nearly full, a backup that failed or is late, sign-in failures out of the ordinary, Claude usage spikes, break-glass use.
- An external check watches the sign-in page and the health endpoint, and alerts the owner.
- The service's logs hold no document content, and are kept for 90 days.
Backups and recovery
| What | How | How often | Kept |
|---|---|---|---|
| Databases (every company's, and the directory) | Continuous archiving of every change, with full and differential backups | Continuous: at most 15 minutes of changes at risk | 30 daily, 12 monthly |
| File store (uploaded documents, images) | File backup | Nightly | 30 daily, 12 monthly |
| Each company's databases on their own | Nightly copies, so one company can be restored without touching any other | Nightly | 30 days |
| Configuration | In version control, and in the backups | On change | History |
- Backups are encrypted on the server before they leave it (Proxmox Backup Server's client-side encryption), with the key kept off the server.
- They go to a separate backup server, which keeps an off-site copy: three copies, two kinds of storage, one off-site.
- A restore is tested every quarter, into a separate machine, and the time it took recorded. Targets: at most 15 minutes of changes lost (RPO) and 4 hours to restore the service (RTO).
Changes
Changes to the service go through version control and the release pipeline (see Application security). Changes to the server's configuration are made through automation playbooks, kept in version control, rather than by hand.
The cloud configuration
The move to Microsoft Azure (regional deployments, managed databases, keys in a key vault, central logging) will be described in its own version of this pack, before any customer is moved.