fix(derp,coord): absorb bugfix/hardening subset of upstream PR #22

[fudio101]

Summary

Started as the narrow DERP TLS-context-leak fix (issue #14); grown, per discussion, into absorbing
the full bugfix/hardening subset of upstream CamM2325/microlink#22
("Support custom (non-Tailscale) control planes end-to-end") — everything in that PR that's useful
regardless of control plane, leaving out the headscale-specific feature work.

DERP (ml_derp.c):

• mbedTLS context leak on every failed ml_derp_connect() (~3-8 KB/attempt) — derp_free_tls_state() + fail_tls goto path
• Region fallback when HomeDERP isn't in the DERPMap (falls back to first usable non-avoid region instead of a dead-end connect)
• TLS 1.2 pin — some relays' Let's Encrypt ECDSA cert trips the ESP-IDF mbedTLS TLS 1.3 signature-algorithm OID path even under VERIFY_NONE
• Return-code checking on every mbedTLS setup call, routed through the same fail_tls cleanup
• BIO rework to a plain blocking f_recv bounded by SO_RCVTIMEO, dropping mbedtls_ssl_conf_read_timeout (avoids a BAD_INPUT_DATA failure the old scheme hit on some relays' TLS 1.3 path)

Coord (ml_coord.c / microlink.h):

• Hostinfo OS field corrected to "esp32" (was "linux")
• New ML_STATE_AUTH_FAILED state — do_register() now actually fails registration on a control-plane Error/MachineAuthorized rejection instead of silently "succeeding" into a broken state (Node absent, confusing downstream "node not found"); same state raised on expired-key-with-no-auth_key
• state_cb now fires on the RECONNECTING transition (it never did before) and isn't clobbered by it while AUTH_FAILED is active

Deliberately excluded (see UPSTREAM_PRS.md's updated "Deliberately skipped" table):

• The 3 headscale/Ionscale-only feature commits from Fix three crash sites from captive-portal DNS failures (mined from cplewes/microlink) #22: build-time CONFIG_ML_CTRL_HOST+TLS control transport (new ml_coord_tls.c), runtime ctrl_host/ctrl_noise_pubkey, streaming_map_fetch — this fork targets Tailscale's own SaaS control plane, no consumer for these.
• 5c8d60c3 (netif->input RX fix) — already absorbed via upstream PR Fix DERP TLS context leak on failed ml_derp_connect() (mined from cplewes/microlink) #20.
• TAI64N wall-clock fix (6a4447c5) — lives in the wireguard_lwip submodule (separate repo fugo101/wireguard-lwip), can't be a commit on this branch. Tracked as a follow-up PR there; FORK_PRS.md issue Adapt dj-oyu/microlink fixes: TAI64N replay-after-reboot, DERP-to-direct re-handshake, LOCK_TCPIP_CORE no-op, DISCO PONG rate-limiting #43 (dj-oyu bundle) updated to note its TAI64N item is superseded by this cleaner source once that lands.

Attribution

Adapted (not literal cherry-picks — base has diverged too far from both sources) from:

• cplewes/microlink 38602ab0, b25b1eee (original leak-fix scaffolding)
• CamM2325/microlink#22 commits 125b5294, fcdc8d95, ae3d4386, 8367c1ed, 9ef10bbb

Related issues

Closes #14
Refs #43 (FORK_PRS.md dj-oyu bundle, TAI64N item superseded — not closed, rest of that bundle is still open)
Refs CamM2325/microlink#22 (upstream, partial absorption — see UPSTREAM_PRS.md)

Test plan

• idf.py fullclean && idf.py build for examples/basic_connect (ESP-IDF wasn't available in the sandbox this PR was written in — please build before merging)
• Manual diff review: region-fallback loop only fires when node_selected is false; every new mbedTLS setup-call failure routes through fail_tls:; all 4 Hostinfo OS occurrences updated (grep -n '"OS", "linux"' returns nothing); ML_STATE_AUTH_FAILED is appended (not inserted) into the enum; COORD_RECONNECTING's ML_CMD_DISCONNECT path unaffected by the new guard
• Hardware re-verification recommended, especially the DERP TLS 1.2 pin / BIO rework and the new AUTH_FAILED state transitions — none of this was hardware-tested in this session

🤖 Generated with Claude Code