Skip to content

Server: the start page and the device lookup served by camex itself (web_listen, nginx auth_request), no sync script #203

Closed

Description

@compot-dev

Problem

The nginx deployment (docs/nginx.md) runs a second service next to camex: camex-nginx-sync --follow, 633 lines of sh run as root. It:

  • asks camex --status --format nginx-map, writes /etc/nginx/camex-map.conf, checks it with nginx -t and runs nginx -s reload whenever the clients change;
  • asks camex --status --format nginx-index every 10 seconds and writes /var/lib/camex-nginx/index.html;
  • keeps its own state next to the map: .applied, .lock, .warnings, .warnings-index, .tmp.$$, .prev.$$.

Both the page and the map already come from camex itself (src/nginxmap.c), so the script only moves them into files. Each reload leaves the old nginx workers alive for up to proxy_read_timeout (300 s) while streams are open, and a server without nginx cannot show its page at all.

Proposal

camex serves the page and the lookup itself; nginx stays the proxy, TLS and authorization in front of it.

camex: web_listen= in [server]

Off by default. Default address when given without one: 127.0.0.1:5801.

Path What Who
GET / the start page, HTML built into the binary (nginx_index_format()) anyone allowed
GET /status.json the data the page refreshes itself from anyone allowed
GET /resolve 204 with X-Camex-Upstream: <ip>:<port>, empty when no such device 127.0.0.1 only (else 404)
  • GET only, request head up to 8 KB, no keep-alive, strict parsing. Control stays on the unix socket: the web side changes nothing.
  • X-Forwarded-Host / X-Forwarded-Proto are trusted only from 127.0.0.1: behind nginx the page links to https://<name>.<domain>/; without nginx it links to the devices' tunnel addresses (http://10.200.0.5:8085/).
  • A build flag leaves the web server out (for cameras, where camex is a client); on by default.

Password without nginx

web_listen Password
127.0.0.1:... (behind nginx) none: nginx asks for it (auth_basic)
a LAN address or 0.0.0.0 required: camex does not start without web_password=
the same, page open on purpose web_password=none written out, with a warning at start
  • web_password= holds an argon2 hash from camex --hash-password, never the password.
  • A checked Authorization header is remembered in memory and compared in constant time, so the page's refresh does not run argon2 every 10 seconds.
  • Failed attempts are limited per IP.
  • Plain HTTP: the docs say this mode is for a LAN; a page on the internet goes through nginx with TLS.

Several servers on one host

Today --status and the script put the clients of every server on the host on one page and one map. Kept:

  • the server with web_listen asks the others through their control sockets (as --status does) and shows them all;
  • web_scope=self: only its own clients, for a server with its own domain and its own web_listen;
  • two servers with web_listen for one domain: a warning at start.

With named networks and --transport both (5.0.0) one daemon covers what used to take several, so this is mostly for existing setups.

nginx: auth_request instead of the map

location / {
    auth_request /_camex;
    auth_request_set $camex_upstream $upstream_http_x_camex_upstream;
    if ($camex_upstream = "") { return 503; }   # -> @offline
    proxy_pass http://$camex_upstream;          # straight to the device, as today
    # WebSocket, timeouts, client_max_body_size 0, @offline: unchanged
}

location = /_camex {
    internal;
    proxy_pass http://127.0.0.1:5801/resolve;
    proxy_pass_request_body off;
    proxy_set_header Content-Length "";
    proxy_set_header X-Camex-Host $host;
}
  • Traffic does not go through camex; per-client ports (web_port) keep working; a device that comes or goes is seen at once, without a reload.
  • auth_request takes any code other than 2xx, 401 and 403 for an error (500), so /resolve always answers 204.
  • The start page server block proxies / and /status.json to camex instead of root /var/lib/camex-nginx.
  • ngx_http_auth_request_module is in the nginx packages of Debian and Ubuntu; other builds need a check.
  • Later, not in this issue: per-device access through /resolve.

Moving to it

install.sh installs the new nginx file and stops installing the script. What an existing server removes by hand is listed in the docs:

  • /usr/local/bin/camex-nginx-sync, /etc/systemd/system/camex-nginx-sync.service (systemctl disable --now camex-nginx-sync first)
  • /etc/nginx/camex-map.conf and camex-map.conf.applied, .lock, .warnings, .warnings-index
  • /var/lib/camex-nginx/

Versions

  • 5.1.0 (after 5.0.0): web_listen, web_password, web_scope, --hash-password, /resolve, the new nginx example and docs; the sync script and --format nginx-map / nginx-index deprecated but still there.
  • 6.0.0: the script and the two formats removed.

Files

Now After
examples/camex-nginx-sync.sh, .service deprecated in 5.1.0, removed in 6.0.0
examples/nginx-camex.conf auth_request, no map include, page proxied to camex
install.sh no script, no page directory
docs/nginx.md "Sync script" replaced by web_listen; the removal list
docker/test-e2e-nginx-map.sh a new e2e test with nginx and auth_request
--format nginx-map / nginx-index deprecated in 5.1.0, removed in 6.0.0
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    deferredPostponed, not planned for nowenhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions