Phase 8D - massive refactor and admin functions / admin shell in

webinterface. Memory and cross-origin problems.

docs/agent/design-decisions.md

@@ -34,13 +34,13 @@ Only constraints supported by implementation or current project documentation be
 
 ## Resource IDs are generation-safe
 
-**Decision:** Broker clients, SSH/WebSocket slots, queued admin work, and user principals carry generations or random stable IDs to reject stale references and slot reuse.
+**Decision:** Broker clients, SSH/WebSocket slots, queued admin work, user principals/accounts, Wi-Fi working configuration, and HTTPS lifecycle intent carry domain-specific generations or random stable IDs to reject stale references, slot reuse, and lost updates.
 
-**Rationale/evidence:** Broker IDs encode slot generation; transports track slot generations; admin tokens include session/slot generation; user principal currentness includes account ID and authentication generation.
+**Rationale/evidence:** Broker IDs encode slot generation; transports track slot generations; admin tokens include session/slot generation; user principal currentness includes account ID and authentication generation; user/Wi-Fi web editors carry optimistic generations; HTTPS snapshots expose lifecycle generation.
 
-**Consequence for future changes:** Preserve transport-slot generations and account-authentication generations as distinct concepts. Validate tokens immediately before side effects and discard late work after disconnect/reuse/revocation.
+**Consequence for future changes:** Preserve transport-slot, account-authentication/database, Wi-Fi working-config, and HTTPS lifecycle generations as distinct concepts. Validate tokens immediately before side effects and discard late work after disconnect/reuse/revocation. Existing-account web mutations must also compare stable user ID so deletion/recreation of the same username cannot retarget stale work.
 
-**Relevant files:** `src/session_broker.{h,c}`, `src/ssh_transport.c`, `src/web_serial_transport.c`, `src/admin_ssh_console.c`, `src/user_database.{h,c}`
+**Relevant files:** `src/session_broker.{h,c}`, `src/ssh_transport.c`, `src/web_serial_transport.c`, `src/admin_ssh_console.c`, `src/user_database.{h,c}`, `src/user_admin_service.{h,c}`, `src/wifi_manager.{h,c}`, `src/web_server.{h,c}`
 
 ## UART0 is the physical recovery authority
 
@@ -92,6 +92,76 @@ Only constraints supported by implementation or current project documentation be
 
 **Relevant files:** `src/user_database.{h,c}`, `src/user_console.c`, `src/web_server.c`, `src/web_serial_transport.c`, `src/ssh_transport.c`
 
+## Browser authentication uses bounded explicit sessions
+
+**Decision:** Browser access uses a same-origin login/logout flow and a fixed RAM session table rather than HTTP Basic. Raw opaque tokens exist only in host-only secure cookies; firmware storage retains token digests, copied principals, monotonic expiry, and generation-safe slot identity. State-changing requests require a session-bound CSRF token and exact Origin validation.
+
+**Rationale/evidence:** Explicit logout and account switching cannot reliably invalidate browser-managed HTTP Basic credentials. Exact browser-session references also allow logout of one session without revoking another session for the same account.
+
+**Consequence for future changes:** Preserve digest-only storage, the two-per-account/eight-global capacity, exact-session WebSocket binding, and current-principal checks. Do not expose tokens, CSRF values, ticket values, or internal references in logs/snapshots. New mutation endpoints must use the body-backed in-place URL-form parser's 512-byte/10-unique-field bounds, parse closed schemas, and revalidate the exact admin session immediately before typed side effects. Browser failure/close paths must clear entered and generated secret material rather than replaying it after a reload.
+
+**Relevant files:** `src/web_session.{h,c}`, `src/web_server.c`, `src/web_serial_transport.c`, `src/web_admin_transport.c`
+
+## Browser admin shell is separate from browser serial
+
+**Decision:** An administrator may keep the browser serial WebSocket alive while independently opening one admin-only WebSocket frontend for the canonical command dispatcher. Terminal switching changes only browser visibility and focus; the admin route never becomes a broker client.
+
+**Rationale/evidence:** This preserves a browser-held writer lease while giving full canonical administrative command parity without a generic HTTP command endpoint.
+
+**Consequence for future changes:** Do not multiplex admin command bytes into `/ws/serial`, and do not close or release the serial route as a side effect of mode switching, settings navigation, or popover display. HTTPD remains the owner of WebSocket send/close calls; the web-admin task only queues bounded work. Self-affecting HTTPS commands must use deferred drain control.
+
+**Relevant files:** `src/web_admin_transport.{h,c}`, `src/admin_ssh_console.{h,c}`, `src/web_ui.c`, `src/web_server.c`
+
+## Browser writer transfer is atomic and generation-safe
+
+**Decision:** Guided writer assignment compares the expected current writer and validates the exact generation-safe target under the broker mutex before making one atomic ownership change.
+
+**Rationale/evidence:** A browser dialog can become stale while open. Unconditional force assignment could overwrite a newer legitimate lease or release ownership when its target disconnected.
+
+**Consequence for future changes:** Use `session_broker_compare_exchange_writer()` for stale UI/API transfers. Opening or hovering a writer control must never mutate ownership, and target/current conflicts must leave the current lease unchanged.
+
+**Relevant files:** `src/session_broker.{h,c}`, `src/web_server.c`, `src/web_ui.c`
+
+## User mutations have one serialized typed boundary
+
+**Decision:** `user_admin_service` owns typed account/password/role/key mutations for console and web callers. Its recursive `admin_command_gate` critical region includes the optimistic snapshot check and database commit; a committed mutation is followed by independent best-effort web and SSH revocation attempts.
+
+**Rationale/evidence:** Browser requests can race one another and canonical shell commands. Database generation plus stable user ID reject stale editors and username delete/recreate races, while the shared gate prevents caller-specific check-then-mutate interleaving. Revocation cannot be made atomic with the NVS commit, so principal currentness remains authoritative.
+
+**Consequence for future changes:** Route new ordinary user mutations through this service instead of calling `user_database` directly. Do not roll back or report a committed mutation as failed solely because a transport notification failed. Preserve final-admin checks, remote self-generated-password restrictions, secret wiping, and generation/user-ID conflict reporting.
+
+**Relevant files:** `src/user_admin_service.{h,c}`, `src/user_database.{h,c}`, `src/user_console.c`, `src/admin_command_gate.{h,c}`, `src/web_server.c`, `src/ssh_transport.c`
+
+## Display configuration writers share the administration gate
+
+**Decision:** Typed browser display Apply/Save/Load/Defaults/Reset operations and console display-writer commands serialize through the recursive `admin_command_gate` for the complete RAM and persistence operation.
+
+**Rationale/evidence:** Browser handlers and canonical console frontends can mutate the same local-UI working configuration and NVS record concurrently. Serializing only individual lower-level calls could allow interleaved apply/save/load/reset sequences and inconsistent final state.
+
+**Consequence for future changes:** Keep all new display configuration writers under the same gate, including any read-modify-write and rollback sequence. Do not hold the gate for read-only status or rendering work, and do not conflate it with the display framebuffer/I2C mutex.
+
+**Relevant files:** `src/web_server.c`, `src/local_ui_console.c`, `src/local_ui_config.{h,c}`, `src/local_status_ui.{h,c}`, `src/admin_command_gate.{h,c}`
+
+## Guided Wi-Fi editing uses exact-generation compare-and-swap
+
+**Decision:** Browser Wi-Fi reads return configuration metadata plus `secret_set` flags, never PSKs. Each typed edit applies a complete validated working-config copy only if its expected nonzero generation remains current, and Save persists exactly the expected generation under the same writer serialization.
+
+**Rationale/evidence:** Multiple browser editors, console changes, and lifecycle controls can update RAM configuration concurrently. A conventional read/modify/write or copy-then-save sequence could overwrite a newer secret or persist a generation the user never reviewed.
+
+**Consequence for future changes:** Keep credential-bearing copies tightly scoped and wiped. Add config writers under the writer mutex and advance generation without wraparound; mismatch or exhaustion must fail closed. Stale browser forms must reload without replay and clear entered secrets.
+
+**Relevant files:** `src/wifi_manager.{h,c}`, `src/wifi_config.{h,c}`, `src/wifi_console.c`, `src/web_server.c`, `src/web_ui.c`
+
+## HTTPS lifecycle and post-material refresh are serialized
+
+**Decision:** HTTPS start, stop, and TLS refresh share a lifecycle mutex and generation-tagged desired-running intent. Certificate/material replacement always proceeds to a TLS refresh. Teardown disables new admin-transport HTTPD calls, tracks calls already in progress, retains a server whose stop failed, and keeps incomplete post-stop finalization pending for retry before another start.
+
+**Rationale/evidence:** Browser-shell commands can tear down their own transport while console commands race a restart or replace persisted TLS material. HTTPD-owned queued work must finish before its handle or the admin transport's static state can be reused.
+
+**Consequence for future changes:** Do not start a second server around a retained handle, bypass lifecycle serialization, or make post-material refresh optional after persistence. A newer explicit lifecycle intent must win over an older refresh. Finalize timed-out detach state only after HTTPD destruction, and retry a failed finalizer before attaching a replacement server.
+
+**Relevant files:** `src/web_server.{h,c}`, `src/web_console.c`, `src/web_admin_transport.{h,c}`, `src/admin_ssh_console.{h,c}`
+
 ## Security material and configuration use bounded, versioned NVS records
 
 **Decision:** Application settings, users, and identities use separate fixed/versioned NVS blobs. Serial, Wi-Fi, mDNS-hostname, and local-UI working edits are RAM-only until explicitly saved. User mutations and HTTPS/SSH identity changes commit directly as part of the operation. Invalid ordinary configuration generally selects RAM defaults without erasing storage; malformed security material fails closed and needs explicit reset.