Files
inkwell/docs/public-hosting.md
bvandeusenandClaude Opus 5.5 7eacd0569c
CI & Build / Python lint (push) Successful in 2s
CI & Build / Build now, or wait for Android? (push) Successful in 2s
Android / Build, or is the channel already serving this? (push) Successful in 3s
Desktop (Tauri) / Build, or is the channel already serving this? (push) Successful in 2s
CI & Build / Web typecheck and unit tests (push) Successful in 8s
CI & Build / Python tests (push) Successful in 10s
CI & Build / integration (push) Failing after 1m21s
CI & Build / Build & push image (push) Skipped
Desktop (Tauri) / Web tests, clippy, Rust tests and rustfmt (push) Successful in 2m47s
Desktop (Tauri) / Windows installer (cross-compiled) (push) Successful in 2m48s
Desktop (Tauri) / Tauri desktop (Linux) (push) Successful in 3m39s
Desktop (Tauri) / Update manifest (push) Successful in 3s
Android / Kotlin + Rust (APK) (push) Successful in 8m37s
all: delete the code nothing calls
From the audit (#5178). Each was unreachable from every client:

- Checklist add-item and delete-item: REST POST /items and DELETE /items/<id>,
  the Tauri commands, the store, rest and local adapters, the core's
  add_item/delete_item, set_item_text and remove_item, and the FFI exports.
  Adding, rewording and removing an item are body edits in every editor. The
  checked toggle stays, and its rewriter is simpler without the drop branch.
- Manual unfurl: POST /unfurl and its adapters. Previews arrive in the
  background after a save (unfurl_queue).
- The /api/config `android_client` key, android_release() and the
  APK_NAME/MANIFEST_NAME aliases. Phones poll /api/client/android.
- users.email_verified and users.avatar_path (migration 0037). Nothing set
  the first or read the second; the SMTP reset never checked verification.
- derive::extract_tags (only tests used it; the shared fixture now runs
  through extract_tag_spans), the unused check and link icons, and the
  unused editor_add_item string.
- The blob scheme is renamed tsblob -> inkblob. URLs are built as notes are
  read, so nothing stored carries the old one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-07 19:00:44 -04:00

168 lines
9.2 KiB
Markdown

# Putting Inkwell on the public internet
Inkwell is built to run on a LAN and works fine there with no ceremony. Exposing
it changes the threat model: anyone can now reach the login form, and any account is
one guessed password away from someone's whole note history.
This is what the app does about that on its own, and the four things it cannot do for
you.
## Do these five things first
**1. Check registration is closed.** On a fresh instance this now takes care of
itself: the first account created becomes the admin *and* closes registration behind
it, so there is no window between "my account exists" and "I remembered to turn it
off". A brand-new instance is never locked out of itself, and never left open either.
**Instances that predate this still need one manual flip.** The close fires when the
first account is created, so a server whose admin already existed keeps whatever
`allow_registration` was set to — which was **on** by default. Check **Settings →
Access → Allow new registrations** before exposing an instance you have been running
on a LAN.
To let someone else in, send them an invite: **Settings → Invites** makes a link that
works once, expires (a week by default), and can be pinned to one email address. It
lets that one person register while registration stays closed.
**2. Terminate TLS in front of it, and forward the scheme.** The app marks the
session cookie `Secure` and sends HSTS only when it can tell the request arrived over
HTTPS. It looks at `X-Forwarded-Proto`, so the proxy has to set it:
```
# Traefik does this automatically. For nginx:
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
```
Without that header the app assumes plain HTTP and leaves the cookie unmarked — the
conservative choice, since forcing `Secure` on an HTTP install stops the browser from
ever sending the cookie back and silently breaks login.
Once a browser has seen HSTS from your hostname it will refuse plain HTTP there for a
year, even if the header stops. That is the point of it, but it is worth knowing
before you put a hostname behind TLS temporarily.
**3. Tell it how many proxies are in front of it.** **Settings → Security → Trusted
proxy hops**, which defaults to `1` — one reverse proxy terminating TLS. Behind a CDN
as well (Cloudflare in front of your proxy) set it to `2`. It applies immediately; no
restart.
This decides which entry of `X-Forwarded-For` is believed, and it is a security
setting rather than a preference. The header grows left to right as a request
traverses, so the rightmost entries are the ones your own infrastructure wrote and
anything a caller forged sits to the left of them. Counting in from the right by the
number of proxies you actually run means a forged prefix can never be selected. Set it
too HIGH and it starts trusting entries no proxy of yours wrote; too low and several
callers share one rate-limit bucket, which is merely inconvenient.
**4. Stop reaching the app except through the proxy.** The default compose binds
`0.0.0.0:5000` so LAN clients can reach it directly — which is right for a LAN install
and wrong the moment there is a proxy in front, because it leaves a second way in that
bypasses everything the proxy does.
Which fix depends on where your proxy runs:
- **Proxy in Docker** (Traefik discovering the container, an nginx container): delete
the `ports:` block from `docker-compose.yml`. The proxy reaches the app over the
compose network; no published port is needed at all, and this is the safest of the
two because there is no host port to reach even from the host.
- **Proxy on the host**: set `INKWELL_BIND=127.0.0.1` in `.env`, so the port
exists but only the host itself can use it.
To check which you have: `docker compose ps` shows the published ports, and
`curl http://<your-lan-ip>:5000/api/health` from another machine tells you whether the
app is still answering around the proxy. It should not be.
**5. Have a backup that includes the files.** Attachments are files on the
`inkwell-data` volume, not rows — a `pg_dump` restores notes whose images are all
gone. Back up both:
```
docker compose exec -T db pg_dump -U inkwell inkwell > notes.sql
docker run --rm -v inkwell-data:/d -v "$PWD":/out alpine tar czf /out/media.tgz -C /d .
```
## What the app already does
- **The credential endpoints are throttled.** `/api/auth/login`, `/api/auth/register`
and `/api/auth/device-login` count attempts against both the account and the calling
address, and answer `429` with a `Retry-After` once either is over budget. The
numbers live in **Settings → Security** — ten failed sign-ins per account per
fifteen minutes and five sign-ups per address per hour by default — and a change
applies to the next attempt rather than the next deploy. The account-keyed limit is the one that holds when the address is forged.
Checked *before* the password is verified, so a throttled attempt costs no bcrypt:
hashing is deliberately slow, and an unauthenticated caller who can trigger it
without limit has a CPU-exhaustion primitive as well as a guessing one.
- **A failed sign-in takes the same time whether or not the account exists.** No
timing oracle for which emails are registered here.
- **Every response carries a CSP** with `script-src 'self'`, `object-src 'none'` and
`frame-ancestors 'none'`, plus `nosniff`, a referrer policy and a permissions
policy. The app has no inline or third-party scripts, so this costs nothing.
- **Link unfurling is SSRF-hardened.** Every hop is resolved and every resolved
address must be publicly routable before a socket is opened, and the connection is
made to the vetted IP so a rebind between check and connect cannot slip through. A
note containing `http://192.168.1.1/` cannot make your server probe your network.
- **Attachments never render inline unless they are a known raster image.** Anything
else — an SVG, an HTML file — is served `Content-Disposition: attachment`, so a file
on a note shared with you can't run script in your session.
- **Session cookies are `HttpOnly` and `SameSite=Lax`**, which is also what stands in
for CSRF protection: a `Lax` cookie is not sent on a cross-site POST.
## Email and forgotten passwords
A forgotten password can be reset two ways. Either way the link works once, within
an hour, and using it signs the account out everywhere and unlinks its apps.
- **By an admin, always.** Settings → People → Reset password makes a link, and the
admin hands it over.
- **By the person, once email is set up.** Fill in Settings → Email (SMTP server,
port, encryption, sign-in and a from address) and Settings → General → Public
address, the URL people reach this server at. Save, then press **Send test email**,
which mails you with exactly the settings a reset will use. From then on the sign-in
screen offers **Forgot password?**
The public address is required because the emailed link has to point somewhere the
server can trust. Built from the request instead, a forged `Host` header would mail
someone a reset link to a site the forger controls.
The SMTP password is kept in the database and never sent back to the browser; the
field shows only that one is saved. The forgot-password page answers the same way,
at the same speed, whether or not an address has an account, and
`reset_emails_per_account` (Settings → Security) caps how many reset emails one
address can be sent.
## What it does not do
Know these before you decide who gets an account.
- **No email verification.** An account's address is whatever was typed when it
registered, and email is used only for password resets (above), which go to that
address. Nothing changes an address afterwards, so someone who registered with a mistyped
one can't reset by email; an admin can issue them a reset link instead.
- **An admin who forgets their own password**, with email off and no other admin,
still needs a hand on the database.
- **No second factor.** A password is the whole of it.
- **No per-user storage quota.** Any account can upload attachments until the volume
is full. `max_attachment_mb` caps a single file, not a total.
- **No audit TABLE.** Credential events — sign-ins, failures, throttle trips, new
accounts, device tokens issued — are written to the application log and readable
with `docker compose logs app`, which is enough to see whether anyone is knocking.
They are not queryable, not retained beyond the container's log rotation, and not
attributable after the fact.
None of these are hard blockers for an instance whose accounts are you and people you
know. They are the reason not to hand out open registration to strangers.
## The Android client
The app allows plain HTTP so a self-hosted server on a LAN is usable at all — Android
blocks cleartext by default from API 28, and `http://192.168.1.10:8000` is exactly the
case Inkwell is built for. Over the public internet, link the phone to the
**HTTPS** hostname. The sync screen shows a warning before any credential field
whenever the address it probed was `http://`; on a public network that warning means
what it says.
The APK the server hands out is signed with the project release key, and the in-app
updater installs over the existing app only because the signature matches. A build
from anywhere else will not install over it.