Replace Web Basic Auth With Cookie Sessions

Add bounded login challenges, CSRF/origin enforcement, logout, and
session-bound WebSocket admission. Isolate private HTTPD access behind a
version-guarded adapter and add focused host coverage. Also let empty
admin
SSH input reach the normal console handler.
This commit is contained in:
2026-09-05 23:55:05 +02:00
parent 4435a7fddd
commit 5a609fa40b
36 changed files with 1940 additions and 360 deletions
+17
View File
@@ -0,0 +1,17 @@
# Cookie authentication and HTTPD adapter host checks
Run from the project root:
```sh
python3 tests/web_cookie_auth/run.py
```
Requires Python 3, a C11 compiler (`cc`), OpenSSL headers/libcrypto, and the pinned ESP-IDF source installation. The runner uses `IDF_PATH` when set, otherwise `~/.platformio/packages/framework-espidf`. It writes only an automatically removed temporary directory. No network, device, pip/npm packages or server is needed. Do not disable C assertions.
The runner compiles production `web_cookie_auth`, `web_session_store`, `web_auth_parse` and `web_httpd_adapter` with bounded HTTPD/database/time/RNG doubles. It also executes the session-store public API suite. The installed IDF header getters, append-only response-header setter and right-aligned pending-data reader are extracted verbatim and compiled into the harness.
Coverage includes challenge reuse/consumption/expiry, capacities without eviction, global throttle, fragmented login bodies, secure cookie attributes and two simultaneous Set-Cookie fields, session-specific logout, duplicate fields/cookies, Origin/CSRF/method/Fetch Metadata rejection, Basic denial, currentness, stop/login and failure paths, six-header login budget, upgrade-state installation, and request cleanup preserving all 0128 pending lengths through partial reads.
This is **not** the full IDF parser/dispatcher, real handshake/TLS/socket, browser, multicore task or hardware test. The private struct doubles do not prove binary layout; firmware compilation uses the actual pinned headers, and the version guard requires a new audit on SDK changes. Handshake sending and transport revocation are doubled. Actual on-wire pre-101 rejection, frame routing, pipelining/early bytes, cookie/CSP/browser recovery and loaded expiry latency remain M1 target gates. No sanitizer or runtime memory-reserve result is implied.
See `docs/phase8d3_implementation.md` for source verification, other suite commands, build accounting and the target checklist.