A VPN's synthetic DNS answer (fake-IP 198.18.0.0/15, 192.0.0.69) now works for a host name. An IP from the server, in any numeric form, must be public; dials to the client itself are closed.
A TUN-mode VPN with fake-IP DNS answers names from 198.18.0.0/15 or 2001:2::/48, which the LAN guard refused, so every dial failed. Refused hosts now warn once.
Basic stays first and Control D last; the rest follow their config key order. Geofence and proxy_map lookups go through pick_proxy and skip a provider that fails.
A value starting with - or @ applies to its region (fr: -datacamp); a value with @ after a server is a whole query (fr: fr-bod@195.86.24.103). A selector in --proxy still wins.
On NordVPN, Windscribe and Surfshark, `@<name>` keeps and `@-<name>` skips a hosting network; `@<ip>` and `@-<ip>` pick or skip one server. The connect line names the server, IP and network. Surfshark pins an IP through a loopback TLS port.
A reverse tunnel over the remote connection: the server's service traffic leaves through the client's own network. Only the client dials. Adds the tunnel-test command.
dl and search keep the typed --proxy value, so --proxy controld:ca --remote sends the server the Control D resolver, not this machine's forwarder. The Control D log names the location and masks the resolver ID in place of 127.0.0.1.
--proxy now overrides HTTP(S)_PROXY and the Windows system proxy while keeping the env CA bundle. Core fallback sessions use the run's proxy, and a warn-only audit hook names service code that sends a request or opens a raw socket outside the proxy; a session with proxy_exempt = True opts out.
A bare query skips a failing provider and errors when none answers; an unknown prefix such as us:seattle fails with a hint. The REST path now loads WindscribeVPN.
Basic, NordVPN, Windscribe, Gluetun and Hola no longer raise on a query they cannot serve. Hola parses stdout only and reports a ban; NordVPN cities use the API filter; uk maps to gb.
Runs a loopback forwarder per region-named profile (unshackle-<region>), so
--proxy and --proxy-download can exit in different countries
Thanks to Orni_ for the original code and the test account.
Co-authored-by: AntCardFr <24798811+AntCardFR@users.noreply.github.com>
Manifest, licence and auth stay on --proxy; track, subtitle and attachment downloads use the second proxy. Also exposed as proxy_download on the REST download job with the same server_proxy gate. The download session now ignores HTTP(S)_PROXY and copies rnet cookies set by name.
A dead proxy, or one that exits from this machine's IP, now fails with INVALID_PROXY instead of retrying every request; remote session create also builds the service off the event loop so a stuck check cannot block the server.
Swap cookie/PKCE auth for the OAuth 2.0 device authorization grant with a cached rotating refresh token. Fix the connection-token key (`cat`) and move the cache to {cache}/vpn/ to match ProtonVPN.
BREAKING CHANGE: cookie-file auth removed; run once with `enable: true` to approve a device code.
imSp4rky