added code-server

README.md

@@ -38,6 +38,7 @@ handled separately — see [Exposing services](#exposing-services).
 |---------|------|-------|
 | [audiobookshelf](audiobookshelf/) | 13378 | Audiobooks and podcasts, with per-user progress sync |
 | [cloudflare-tunnel](cloudflare-tunnel/) | — | Outbound connector. Exposes other services without opening a port |
+| [codeserver](codeserver/) | 8443 | VS Code in a browser. Effectively a remote shell — read its README before exposing |
 | [jellyfin](jellyfin/) | 8096 | Media streaming. Also 7359/udp for auto-discovery |
 | [openwebui](openwebui/) | 3033 | LLM chat UI. One `OWUI_TAG` switches between CPU, CUDA and bundled-Ollama |
 | [overleaf](overleaf/) | 8020 | LaTeX editor. Four containers: app, Mongo (replica set), Redis, init |

codeserver/.env.example

@@ -0,0 +1,57 @@
+# Copy to .env and adjust. .env is git-ignored; this file is not.
+
+# ─── Required ───────────────────────────────────────────────────────
+# Web UI password. There is no default and no "first run" setup: if neither
+# PASSWORD nor HASHED_PASSWORD is set, code-server serves an editor and a
+# terminal with NO authentication whatsoever.
+PASSWORD=
+
+# Instead of the plaintext above, you can store only a hash. It overrides
+# PASSWORD when set. Generate one with:
+#
+#   echo -n "yourpassword" | npx argon2-cli -e
+#
+# Then uncomment HASHED_PASSWORD in compose.yaml.
+#HASHED_PASSWORD=
+
+# ─── Sudo inside the container ──────────────────────────────────────
+# Set this if you want to install things from the built-in terminal. The image
+# is deliberately minimal -- no python, no pip, no compilers -- so anything
+# beyond editing needs apt:
+#
+#   sudo apt update && sudo apt install -y python3-pip
+#
+# Leaving it empty is safe and simply disables sudo; the init script tests for
+# a non-empty value, so empty is not an enabled blank password. It can be the
+# same as PASSWORD or different -- it only guards `sudo` inside the container.
+#
+# Be aware of what it grants: root in the container. That is a container, not
+# your host, but it is still root over anything you have mounted.
+SUDO_PASSWORD=
+
+# Store a hash instead of the plaintext. Takes priority over SUDO_PASSWORD.
+# Format is $type$salt$hash. Uncomment the matching line in compose.yaml.
+#SUDO_PASSWORD_HASH=
+
+# ─── Identity ───────────────────────────────────────────────────────
+# Files you create in the editor are owned by this user. `id -u` / `id -g`.
+# Defaulted to 1000/1000 in compose.yaml.
+#PUID=1000
+#PGID=1000
+#TZ=Europe/Paris
+
+# ─── Paths ──────────────────────────────────────────────────────────
+# Settings, installed extensions, and the default workspace.
+#CONFIG_PATH=./config
+
+# Optional: existing code on the host to expose in the editor. Uncomment the
+# matching volume line in compose.yaml too.
+#PROJECTS_PATH=/srv/projects
+
+# ─── Ports and proxying ─────────────────────────────────────────────
+#PORT=8443
+#DEFAULT_WORKSPACE=/config/workspace
+
+# Only for subdomain port-forwarding (3000.code.example.com). Needs wildcard
+# DNS and a wildcard certificate; skip it if you don't need that.
+#PROXY_DOMAIN=code.example.com

codeserver/README.md

@@ -0,0 +1,112 @@
+# code-server
+
+VS Code in a browser. Useful for editing on the NAS from a laptop, a tablet, or
+any machine you don't want to set a toolchain up on.
+
+## Quick start
+
+```bash
+cp .env.example .env
+$EDITOR .env            # PASSWORD is required
+docker compose up -d
+```
+
+Open `http://<host>:8443`. The password from `.env` is the only credential.
+
+## Ports
+
+| Port | Proto | Purpose |
+|------|-------|---------|
+| 8443 | http | Editor UI (plain HTTP — put TLS in front of it) |
+
+## Read this before exposing it
+
+code-server is not a document editor with a terminal bolted on. It **is** a
+shell, with a file browser, running as `PUID`/`PGID`, with network access to
+everything the container can reach. Treat exposing it exactly as you'd treat
+exposing SSH.
+
+Three consequences worth being deliberate about:
+
+**`PASSWORD` is required here** (`${PASSWORD:?…}`), which is a deviation from
+upstream's example. Their compose marks it `#optional`, and it is — but the
+docs also say: *"If PASSWORD or HASHED_PASSWORD is not provided, there will be
+no auth."* Optional-and-silently-open is the wrong default for something that
+hands out a shell, so this stack refuses to start without one.
+
+**`SUDO_PASSWORD` is how you install anything.** The image is deliberately
+minimal — no python, no pip, no compilers — so `sudo apt install python3-pip`
+is the normal first move, and that needs sudo enabled. It grants root *in the
+container*, which is a real boundary but not your host. Leave it empty if you
+only ever edit; the init script tests for a non-empty value, so empty means
+disabled rather than an enabled blank password.
+
+**Whatever you mount is executable, not just editable.** The commented
+`PROJECTS_PATH` line is genuinely convenient and genuinely a decision — mount
+your code, not your whole home directory. Never mount `/var/run/docker.sock`
+here; that converts a container password into root on the host.
+
+## Why it looks like this
+
+**LinuxServer image**, so the usual `PUID`/`PGID`/`TZ` contract applies and
+files you create are owned by you rather than root. Same reasoning as
+[jellyfin](../jellyfin/) — one dialect across the stacks.
+
+**`HASHED_PASSWORD` is offered but commented.** `.env` is git-ignored, so a
+plaintext password there is not exposed, but the hash means the credential
+isn't sitting in a readable file at all. Generate one with
+`echo -n "yourpassword" | npx argon2-cli -e`. It overrides `PASSWORD` when set.
+
+**`PROXY_DOMAIN` is commented out.** It only matters if you want forwarded
+ports served as subdomains — port 3000 as `3000.code.example.com` — which needs
+a wildcard DNS record and a wildcard certificate. Without it, forwarded ports
+still work through the editor's own proxy path.
+
+**`DEFAULT_WORKSPACE` is set** so the editor opens somewhere useful instead of
+an empty file picker on every visit.
+
+## What survives a rebuild, and what doesn't
+
+`/config` is the app user's home directory — `abc:x:911:1001::/config` — and
+it's the only persisted path. That single fact decides how you should install
+things:
+
+| Installed with | Lands in | Survives `docker compose up` after a pull? |
+|---|---|---|
+| `sudo apt install …` | `/usr/…` | **No** — gone with the old container |
+| `pip install --user …` | `/config/.local` | Yes |
+| VS Code extensions | `/config` | Yes |
+| Anything in your workspace | `/config/workspace` | Yes |
+
+So `sudo apt install python3-pip` has to be repeated after every image update,
+while the packages you then `pip install --user` stay put. Prefer `--user`
+installs, and keep a note of the apt packages you need so you can replay them.
+
+If a toolchain matters enough that re-running apt annoys you, build a small
+image `FROM lscr.io/linuxserver/code-server` with the packages baked in.
+
+## Gotchas
+
+- **Extensions live in `/config`.** They're part of what to back up, and
+  they're why the config directory grows.
+- **The VS Code Marketplace isn't available.** code-server uses Open VSX, so a
+  few Microsoft-licensed extensions are missing. Most have equivalents.
+- **Port forwarding is per-session.** Ports you open inside the editor are
+  reachable through it, not published on the host.
+- **A container password is a weaker boundary than it looks.** Anyone who logs
+  in can reach every service on your Docker network directly by name.
+
+## Exposing it
+
+- [Reverse proxy with Caddy](../docs/reverse-proxy.md) — the minimum. Never
+  serve this over plain HTTP off your LAN; the password crosses the wire.
+- [Single sign-on with Authentik](../docs/authentik-sso.md) — a good fit here.
+  Forward auth in front of code-server means an attacker needs to get past
+  Authentik before reaching a login form that leads to a shell.
+- [Cloudflare Tunnel](../cloudflare-tunnel/) — works, and avoids opening a port.
+  Pair it with Cloudflare Access rather than relying on the password alone.
+
+## Links
+
+- Image docs: <https://docs.linuxserver.io/images/docker-code-server/>
+- Upstream: <https://github.com/coder/code-server>

codeserver/compose.yaml

@@ -0,0 +1,44 @@
+services:
+  code-server:
+    image: lscr.io/linuxserver/code-server:latest
+    container_name: code-server
+    restart: unless-stopped
+
+    environment:
+      - PUID=${PUID:-1000}
+      - PGID=${PGID:-1000}
+      - TZ=${TZ:-Etc/UTC}
+
+      # REQUIRED. With neither PASSWORD nor HASHED_PASSWORD set, code-server
+      # starts with NO authentication -- an editor and a shell, open to anyone
+      # who can reach the port.
+      - PASSWORD=${PASSWORD:?set one - without it code-server has no auth at all}
+
+      # Overrides PASSWORD when set. Use it to keep the plaintext out of .env.
+      #- HASHED_PASSWORD=${HASHED_PASSWORD}
+
+      # Enables `sudo` in the built-in terminal -- needed to apt-install
+      # anything, since the image ships no python/pip/compilers. Empty is safe
+      # and simply leaves sudo disabled: the init script tests `-n`, so an
+      # empty value is not the same as an enabled blank password.
+      - SUDO_PASSWORD=${SUDO_PASSWORD:-}
+      # Takes priority over SUDO_PASSWORD if you'd rather store only a hash.
+      #- SUDO_PASSWORD_HASH=${SUDO_PASSWORD_HASH}
+
+      # Only needed if you serve forwarded ports on subdomains, so that
+      # port 3000 becomes 3000.code.example.com. Requires a wildcard DNS
+      # record and a wildcard certificate.
+      #- PROXY_DOMAIN=${PROXY_DOMAIN}
+
+      - DEFAULT_WORKSPACE=${DEFAULT_WORKSPACE:-/config/workspace}
+      #- PWA_APPNAME=${PWA_APPNAME:-code-server}
+
+    volumes:
+      # Settings, extensions, and the default workspace. Back this up.
+      - ${CONFIG_PATH:-./config}:/config
+      # Code that lives elsewhere on the host. Whatever you mount here is
+      # editable -- and runnable -- from the browser.
+      #- ${PROJECTS_PATH}:/config/workspace/projects
+
+    ports:
+      - "${PORT:-8443}:8443"

overleaf/README.md

@@ -167,11 +167,64 @@ ln -s overleaf.env .env
   `docker compose exec mongo mongosh --eval 'rs.initiate({_id:"overleaf",members:[{_id:0,host:"mongo:27017"}]})'`
 - **No signup page until the first user exists.** Use the `create-user.mjs`
   script above; the web UI won't offer to make one.
+- **Missing LaTeX packages** — see the section below. The short version is
+  `tlmgr --usermode install <pkg>`, not plain `tlmgr install`.
 - **Back up `DATA_PATH`.** `data/mongo` holds every project. `data/overleaf`
   holds uploaded files and compile output.
 - **Upgrading across majors** may need migrations, and Mongo itself can't jump
   several versions at once. Read the release notes rather than pulling blind.
 
+## Installing LaTeX packages so they stay installed
+
+The image ships a reduced TeX Live, so a document can fail on a package that
+exists upstream — `booktabs` is a common first casualty. The obvious fix works
+and then quietly undoes itself:
+
+```bash
+docker exec overleaf tlmgr install booktabs      # works now, gone after an upgrade
+```
+
+TeX Live lives at `/usr/local/texlive/2026/`, **inside the image**. A
+`docker restart` keeps those files; recreating the container after
+`docker compose pull` does not. A document that compiled last month stops
+compiling and nothing tells you why.
+
+The fix is `TEXMFHOME`, which this stack bind-mounts to `${DATA_PATH}/texmf`.
+Install in user mode and packages land on the volume instead:
+
+```bash
+# once, to create the tree
+docker exec overleaf tlmgr init-usertree
+
+# thereafter, for each package
+docker exec overleaf tlmgr --usermode install booktabs
+```
+
+Verify it took:
+
+```bash
+docker exec overleaf kpsewhich booktabs.sty
+#   /root/texmf/tex/latex/booktabs/booktabs.sty   <- persisted
+#   /usr/local/texlive/...                        <- inside the image, will be lost
+```
+
+The path in the output tells you which mode you used.
+
+**Caveats.** User mode handles ordinary style packages cleanly. Packages that
+install fonts or need map-file updates may also want `updmap-user`, and a few
+that expect to modify the main tree won't work this way — install those with a
+custom image instead:
+
+```dockerfile
+FROM sharelatex/sharelatex:6.2.2
+RUN tlmgr install <packages>
+```
+
+That's the more reproducible option in general, since the package list ends up
+in version control. The trade is rebuilding on every upstream release. If you
+stay with user mode, keep a list of what you've installed — the volume survives,
+but a list is what lets you rebuild from nothing.
+
 ## Health checks
 
 The image ships its own:

overleaf/compose.yaml

@@ -18,6 +18,10 @@ services:
 
     volumes:
       - ${DATA_PATH:-./data}/overleaf:/var/lib/overleaf
+      # TEXMFHOME. Packages installed with `tlmgr --usermode install` land here
+      # instead of inside the image, so they survive an image update. See the
+      # README -- without this, every upgrade silently drops them.
+      - ${DATA_PATH:-./data}/texmf:/root/texmf
 
     environment:
       OVERLEAF_APP_NAME: ${OVERLEAF_APP_NAME:-Overleaf}