Files
Discord-Authenticator/README.md
T
DSH Agent b21ade05eb v0.1.3: make the vault survive a power cut, and recover if it doesn't
The vault was being written with `fs::write(tmp)` → `rename(tmp, path)`. The
rename is atomic, so readers never saw a half-written file — but atomicity is
not durability. Nothing fsynced the new bytes before the rename, so on ext4
(`data=ordered`, the default) a crash or hard power-off could commit the
rename while the data it published never landed. The result is precisely the
failure that stopped the container booting: `vault.json` present, zero bytes,
"EOF while parsing a value at line 1 column 0".

Two things made it worse. The write happened on *every* push tick, so a
ten-account vault was rewriting all of its secrets thousands of times a day,
and there was no backup to fall back on — a damaged file meant a crash loop
with the web UI locked behind the service that would not start.

Storage

  * `store_io::write_atomic` fsyncs the new bytes, renames, then fsyncs the
    directory. That ordering is the whole fix; the module docs explain why.
  * Every vault write refreshes `vault.json.bak`, taken *after* the new file is
    safely in place so the two always agree. Backing up first would leave the
    backup a full change behind and quietly lose the most recent edit.
  * `Vault::load` falls back to the backup when `vault.json` is unparseable,
    moves the damaged file to `vault.json.corrupt-<ts>` rather than deleting
    it, and writes the recovered accounts back so the next boot is ordinary.
    With no usable backup it still refuses to start, as before.

Write volume

  * Discord bookkeeping — message ids, last codes, last errors — moves to its
    own `push-state.json`, batched and flushed at most every five seconds, with
    a final flush on shutdown. Codes roll constantly; accounts change only when
    the operator edits them, and keeping the two together was what put the
    secrets in the path of every rotation.
  * `push-state.json` is a cache by construction: losing or corrupting it costs
    a reposted tile, never a secret, so it is never a reason to refuse to boot.

Tests: recovery from a zero-length vault, backup retention, write-back on
recovery, refusal when both files are bad, reload of tile ids, batching, and
that a rotation leaves `vault.json` byte-identical. 143 passing.
2026-09-20 02:30:22 -04:00

387 lines
17 KiB
Markdown

<p align="center">
<img src="docs/brand/icon-256.png" width="128" height="128" alt="Discord Authenticator" />
</p>
<h1 align="center">Discord Authenticator</h1>
<p align="center">
A self-hosted TOTP authenticator whose codes appear in Discord as
<strong>live messages that rewrite themselves</strong> as each code rolls.
</p>
The web UI is what you'd expect from Google or Microsoft Authenticator: a
searchable list of accounts, each with its six digits and a countdown ring.
The part that isn't ordinary is the delivery — if you have Cloudflare, Plex and
Ubisoft in the vault, your Discord channel holds three messages, each one
staying current on its own.
Written in Rust. One static binary in a 27 MB Alpine image (about 8 MB to
pull), no database, no external services beyond Discord itself.
---
## Contents
- [How the live delivery works](#how-the-live-delivery-works)
- [Security: read this before you use it](#security-read-this-before-you-use-it)
- [Running it](#running-it)
- [Using it](#using-it)
- [Configuration](#configuration)
- [Architecture](#architecture)
- [Troubleshooting](#troubleshooting)
- [Development](#development)
---
## How the live delivery works
The obvious approaches don't survive contact with Discord, and it's worth
saying why, because the design is a direct consequence.
**A live-updating image URL doesn't work.** Pointing an embed at an image your
server renders on demand fails twice over. Discord fetches the URL once at post
time and stores a snapshot; clients then render a cached copy from
`media.discordapp.net`, so new bytes at the same URL are never seen. And more
seriously, Discord's crawler must reach that URL anonymously — which would mean
publishing your live TOTP codes on an unauthenticated public endpoint. That
alone disqualifies it.
**An animated image doesn't work either.** Animation playback starts whenever
the client decodes the file, not on a wall clock. A 30-second countdown baked
into frames drifts out of phase with the real window and ends up confidently
displaying the wrong time remaining — worse than showing nothing.
**What does work is editing the message.** Posting to a webhook with
`?wait=true` returns the created message's id; `PATCH .../messages/{id}` then
rewrites that message in place, and every client sees the change at once. So
each account owns one message for its lifetime, and the app rewrites it once
per period.
Two details make this pleasant rather than merely functional:
- **The countdown costs nothing.** Discord renders `<t:1756402860:R>` as
"in 24 seconds" and re-renders it every second *client-side*. So the ticking
countdown needs no requests at all — only the digits do.
- **The digits stay copyable.** The code goes into a fenced code block, which
Discord gives a copy button, containing the bare digits (`482913`, not
`482 913`) so what lands on your clipboard is directly pasteable. The
human-readable grouped form lives on the rendered tile.
Each tile is a WebP rendered by the app — issuer, account, the code at a size
you can read across a room, and a per-issuer accent colour that matches the row
in the web UI.
### Rate limiting
TOTP windows are globally aligned, so every account's code rolls at the same
instant. Left alone, ten accounts would fire a ten-request burst into a bucket
that allows roughly five per second. Every call is therefore serialised through
a limiter that spaces requests about 500 ms apart, spreading ten edits across
five seconds of a thirty-second window, and backing off on `429` using the
`retry_after` from the response body.
Failures are classified rather than blindly retried, because Discord bans an IP
after 10,000 4xx responses in ten minutes:
| Discord says | Meaning | What happens |
|---|---|---|
| `10008` Unknown Message | You deleted the tile | Post a fresh one automatically |
| `10015` Unknown Webhook | Webhook deleted | Stop, and show you the reason |
| `50027` Invalid token | URL rotated | Stop, and show you the reason |
| `50013` Missing permission | Lost channel access | Stop, and show you the reason |
| `429` | Rate limited | Back off, skip the stale edit |
Late edits are dropped rather than queued — if an edit is overdue the code has
already changed, so the current one is sent instead of replaying a backlog.
---
## Security: read this before you use it
**Anyone who can read that Discord channel can read your codes.** That is most
of what a second factor is for. This tool deliberately trades some of that
security for convenience, and you should decide knowingly:
- Use a **private channel in a server only you are in**.
- Treat the **webhook URL itself as a secret** — it is an unauthenticated write
capability for that channel.
- Anything with access to your Discord account — another logged-in device, a
session token, a member of that server — inherits your second factor.
- Consider leaving push **off** for your highest-value accounts. It's a
per-account toggle, so a mixed setup is fine.
What the app does hold up its end on:
- TOTP secrets are encrypted at rest with **XChaCha20-Poly1305**; `vault.json`
contains ciphertext only, and the metadata beside it is deliberately readable.
- `vault.json` is written atomically *and* durably — new bytes are fsynced
before the rename that publishes them, and the directory is fsynced after —
so an unclean shutdown can't leave a truncated file behind. The previous
contents are kept in `vault.json.bak`, and startup falls back to it if
`vault.json` is ever unreadable, moving the damaged file aside as
`vault.json.corrupt-<timestamp>` rather than refusing to boot.
- The admin password is hashed with **Argon2id**.
- The vault key lives in `vault.key` (mode `0600`), or set `DA_MASTER_KEY` to
keep it off the data volume entirely — then a leaked copy of that volume is
useless on its own.
- Everything under `/api/*` requires a session. There is no public read path.
- The UI is fully offline: no CDN, no remote fonts, no external requests. The
only outbound traffic the app ever makes is to Discord.
**Why the key can't just be your password:** the app has to mint codes
unattended to keep the tiles live, so nothing can require a human to type a
passphrase after a restart. The key is therefore readable by the process at
boot. That protects against a leaked backup of the vault file — the realistic
homelab accident — but not against someone who can read the whole data
directory. `DA_MASTER_KEY` closes that second gap if you need it closed.
---
## Running it
### Docker Compose
```bash
docker compose up -d
```
Then open `http://<host>:8710/`.
### Unraid
Copy `deploy/unraid/discord-authenticator.xml` into
`/boot/config/plugins/dockerMan/templates-user/` and add the container from the
Docker tab, or point the Repository field at:
```
gitea.milesward.dev/mward4/discord-authenticator:latest
```
Bridge networking is correct — this makes outbound HTTPS calls and serves one
port. Map `/var/lib/discord-authenticator` to somewhere on `appdata` and **back
that path up**: losing it loses every stored secret.
### What lives in the data directory
| File | Holds | If it's lost |
| --- | --- | --- |
| `vault.json` | Every account, with its secret sealed | Everything — this is the file to back up |
| `vault.json.bak` | The same accounts, as of the last change | Nothing, unless `vault.json` is damaged |
| `vault.key` | The vault key (unless `DA_MASTER_KEY` is set) | Every secret becomes unreadable |
| `push-state.json` | Discord message ids and last-known push errors | Nothing — tiles are reposted on the next tick |
| `auth.json`, `settings.json` | Admin login and webhook config | You redo first-run setup |
`push-state.json` is deliberately separate from the vault. Codes roll every 30
seconds, so the bookkeeping changes constantly while the accounts themselves
change only when you edit them; keeping them together meant every rotation
rewrote all the secrets.
### Building and publishing the image
```bash
GITEA_TOKEN=<token with write:package> ./scripts/build-and-publish-unraid.sh
```
---
## Using it
**First run** asks for a username, a password, and (optionally) your Discord
webhook URL. Get the webhook from Discord: Server Settings → Integrations →
Webhooks → New Webhook → Copy Webhook URL. All three are changeable later from
Settings.
**Adding an account** offers three routes:
- **Upload** a QR code image — PNG, JPEG, WebP, GIF, BMP or TIFF. Screenshots,
photos of a screen, tiny crops and rotated images all work; the decoder walks
a ladder of rescale, contrast and rotation attempts before giving up. (HEIC
is rejected with a note to screenshot instead — iPhone screenshots are PNG.)
If the image holds several QR codes, all of them are read.
- **Camera** scanning, using the browser's native barcode detector where it
exists and falling back to server-side decoding where it doesn't, which
covers Safari and Firefox. Note that browsers only grant camera access over
HTTPS or on `localhost`.
- **Paste** an `otpauth://` link, or type a secret by hand.
### Importing from Google Authenticator
Google Authenticator's **Transfer accounts → Export accounts** produces a QR
code holding *many* accounts at once, not one — it's an
`otpauth-migration://` link wrapping a base64 protobuf. Upload or paste it and
you get a checklist of everything inside, with each entry's algorithm, digit
count and period shown, and an **Add N accounts** button.
Details worth knowing:
- **Large exports are split across several QR codes.** Google shows them as
separate screens. Each one reports its position, so the UI tells you
"Part 1 of 2" and prompts you to upload the rest. Import them in any order.
- **Re-importing is safe.** Anything already in your vault is detected and
pre-unticked, so scanning an export twice adds nothing. Duplicates are
matched on the secret *and* the naming, so deliberately keeping one secret
under two labels still works.
- **Entries that can't be imported are named, not silently dropped** — a
counter-based (HOTP) account or an MD5 one is listed with the reason.
- **Secrets never reach the browser.** The preview returns names and
parameters only; on import the client sends back the link plus the indices
it selected, and the server re-parses it.
Two traps this format sets, both of which silently produce *plausible but
wrong* codes if you get them backwards: the protobuf `secret` field is **raw
bytes**, not base32, and the payload carries **no period field** — every Google
export is 30-second TOTP.
The parser is deliberately forgiving of real-world URIs: it accepts lowercase
and `=`-padded secrets, and where the issuer in the path disagrees with the
`?issuer=` parameter — which Microsoft does — it takes the query parameter
rather than refusing the account.
**In the list**, click any code to copy it. `/` focuses the search box.
The pencil on a row opens **Edit account**. The *Name* is what shows in the
list and on the Discord tile; the *Account* field underneath (usually the
username or email the code belongs to) is left alone unless you change it —
it's editable, since providers sometimes put something unhelpful there. The
same sheet pauses Discord push for that account, or removes it entirely.
A rename reaches Discord within a second or two. It doesn't wait for the code
to roll: the push loop normally skips any account whose current code it has
already sent, so renaming clears that marker to force one refresh.
---
## Configuration
All settings are environment variables (or CLI flags of the same name).
| Variable | Default | Meaning |
|---|---|---|
| `DA_HTTP_ADDR` | `0.0.0.0:8710` | Listen address for UI and API |
| `DA_WORK_DIR` | `/var/lib/discord-authenticator` | Vault, key, settings, admin account |
| `DA_MASTER_KEY` | *(unset)* | Base64 of 32 bytes. Overrides `vault.key`; nothing is written to disk |
| `DA_LOG` | `info,da=info` | `tracing-subscriber` filter |
Generate a master key with `openssl rand -base64 32`. Setting it *after*
secrets exist makes them unreadable — export first if you're migrating.
**Message style** (Settings → Discord) chooses what each tile contains:
- **Embed** — code as text only. One JSON edit per period, nothing uploaded.
The lightest option by a wide margin.
- **Image** — the rendered WebP tile only.
- **Both** *(default)* — tile plus copyable text.
The image styles re-upload a tile on every roll, which is roughly 2,880
attachments per account per day. Harmless, but if you run many accounts and
want the traffic minimal, Embed is the frugal choice.
---
## Architecture
```
crates/
core/ TOTP, vault, crypto, settings, admin auth — no network
(incl. Google Authenticator export parsing)
render/ SVG → WebP tile rendering (resvg + embedded fonts)
discord/ Webhook client, rate limiter, live-tile push engine
webui/ Embedded HTML/CSS/JS, compiled into the binary
http-api/ axum router, sessions, QR decoding
discord-authenticator/ The binary
```
`core` is deliberately network-free so the security-critical parts are testable
without a socket. TOTP is implemented directly over HMAC (about thirty lines)
rather than pulling a crate — the deciding factor was that real `otpauth://`
URIs need the lenient parsing described above, which the available crates
reject, so the normalisation layer would have had to be written anyway.
Everything is pure Rust: no OpenSSL, no `-sys` crates, no C. TLS roots are
compiled in via `webpki-roots`, which is why the runtime image needs no
`ca-certificates` and the binary is genuinely standalone.
The UI follows [Beautiful UI](https://www.beautifului.dev/) — OKLCH colour
ramps, hairline-ring shadows, inset fields against raised buttons, Inter and
JetBrains Mono. Light and dark both ship.
### Testing
143 tests, including:
- The **RFC 6238 test vectors** for SHA-1, SHA-256 and SHA-512.
- Real **QR round-trips** — clean, 1-pixel-per-module, oversized, rotated, JPEG,
and two codes stacked in one screenshot.
- **Google export decoding** driven from real protobuf wire bytes: multi-account
payloads, batch metadata, every algorithm and digit variant, all four base64
spellings, unknown-field forward compatibility, and truncated input. One test
asserts the raw-bytes secret handling against the RFC 6238 vector, because
that is the mistake that would otherwise ship working-looking wrong codes.
- The **live push cycle against a mock Discord**: that the first call carries
`?wait=true`, that later passes PATCH the same message rather than posting
again, that a deleted tile is reposted, that a dead webhook is surfaced
instead of hammered, that concurrent rollovers are paced apart, and that a
rename refreshes the tile immediately without re-posting it or looping.
- **Crash recovery**: that a zero-length `vault.json` is recovered from the
backup, that the damaged file is kept rather than deleted, that the recovered
accounts are written back so the next boot is ordinary, that an unusable
backup still refuses to boot rather than starting empty, and that a code
rotation does not rewrite the file holding the secrets.
```bash
cargo test --workspace
cargo clippy --workspace --all-targets
```
---
## Troubleshooting
### The container won't start: "could not load the vault"
The app refuses to start on an unreadable `vault.json` because the alternative
— booting with an empty vault — invites you to re-add accounts you already have.
Since v0.1.3 it recovers on its own first: if `vault.json.bak` parses, the
accounts are loaded from it, the damaged file is moved to
`vault.json.corrupt-<timestamp>`, and a fresh `vault.json` is written. You will
see one `ERROR` line saying so, then normal startup.
If it still refuses, both files are unreadable and there is nothing to recover
from automatically. Get the file out of the way and restart:
```bash
cd /mnt/user/appdata/discord-authenticator
mv vault.json "vault.json.broken-$(date +%s)"
docker restart Discord-Authenticator
```
The service will start with an empty vault, so re-import your accounts from
your own backups (`otpauth://` links or a Google Authenticator export). Keep
`vault.key` — without it, nothing in the old file can be decrypted even if the
bytes are recovered later.
**If you hit this, upgrading matters more than the workaround.** The cause was
a rename being journalled ahead of the data it published, which is fixed by
fsyncing before the rename; see `crates/core/src/store_io.rs` for the ordering
and why it is the whole trick.
---
## Development
```bash
cargo run -p discord-authenticator -- --work-dir ./data/work --http-addr 127.0.0.1:8710
```
Preview the Discord tile without a webhook:
```bash
cargo run -p da-render --example preview -- /tmp
```
## Licence
MIT OR Apache-2.0. JetBrains Mono is bundled under the SIL Open Font License
1.1 (`crates/render/assets/JetBrainsMono-OFL.txt`).