Skip to content

feat(servers): admit authenticated remote multi-backend clients - #1550

Draft
pascalandr wants to merge 4 commits into
feat/multi-backendfrom
feat/multi-backend-remote-clients-admission
Draft

pascalandr wants to merge 4 commits into
feat/multi-backendfrom
feat/multi-backend-remote-clients-admission

Conversation

@pascalandr

Copy link
Copy Markdown
Contributor

Admit authenticated RC/HTTPS/LAN UI clients to host saved servers behind multiBackend, retaining the closed server protocol and host-only exclusions.
Enforce profile-administration provenance and refuse new secrets over plain LAN HTTP; expose host capabilities in Home/Servers.
Verified: 2,260 CI UI units, 4,743 server tests passed (25 skipped), 140 targeted browser tests, typechecks, inventory and help ×10.

Allow paired Remote Control and declared direct UI clients to use and administer host-owned server profiles behind the strict multiBackend flag. Preserve local callers and the closed server-device protocol, nested-relay exclusions, and host-only preview and memory boundaries.

Centralize channel admission and publish use, management and secret-entry capabilities. Enforce the profile-state mutation boundary before persistence, including owner deletion and reverse-proxy provenance. Refuse new credentials over LAN HTTP and reject credential-bearing probe URLs without exposing their input.

Add red-first production-application admission, relay, persistence and browser regressions using controlled A/B servers with colliding IDs. Validate the three CI UI unit groups (2260 passed), complete server glob at concurrency 4 (4743 passed, 25 skipped), 140 selected browser tests, typechecks, reviewed asynchronous inventory and all ten help locales.
Integrate the development sync, proxy-destination guard, occurrence-scoped views and trimmed UI copy. Resolve the Home conflict by retaining the trimmed copy and capability-based server controls.
Block remote-server passwords and pairing links in input controls and at every credential dispatch when the host refuses secret entry. Preserve host LAN login and secret-free saved-server operations, with actionable states in all ten locales.

Capture remote relay principals before body parsing and bind queued work, upstream IO and streamed publication to auth invalidation. Reuse borrower-aware authentication recovery so ending one principal cannot cancel another. Keep local-loopback authority and independent direct HTTPS devices intact. Apply fixed-route logging even to rejected relay targets so proof queries stay private.

Add real-listener logout, expiry, password-change, device and shared-login regressions plus the six-channel/transport secret-sentinel matrix with actual TLS, TCP forwards and trace Pino capture. Validate 2376 CI UI units, 4776 server tests passed with 25 skipped, 353 focused server cases, 178 browser cases, both reviewer reproductions, all typechecks, unchanged async inventory and ten-language help.
Capture one immutable remote UI principal before body parsing for relay, browse, connect, probe, inventory and profile administration. Carry cancellation through queue waits and IO, check before dispatch and publication, and retain durable synchronous effects. Keep originless public HTTP proxy requests in their remote class and refuse unsupported control WebSocket upgrades without disturbing content proxies.

Use borrower-aware authentication for non-local browsing so cancellation cannot fail another principal's bootstrap. Preserve private per-connect candidates and profile-owned SSH readiness. Emit only static route and completion metadata, and sanitize framework URL errors instead of reflecting proofs. Document the complete operation-lifetime contract.

Extend real listener regressions and sentinel scans to padded and unpadded base64/base64url, URL/form encodings, captured network data and browser storage. Validate 2376 CI UI units, 4816 server tests passed with 25 skipped at concurrency four, 390 remote-family cases, 179 browser cases, 65 focused operation cases, all typechecks, reviewed inventory and help. Execute all three unchanged reviewer scripts through a checkout-local resolver; positive regressions verify the corrected outcomes.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant