Alan-Jowett / sonde / 24640537270

82%

push

github

modem: fix Android LESC NC pairing via encryption-state polling fallback (#763)

* modem: fix Android LESC NC pairing via encryption-state polling fallback

When Android uses LESC Numeric Comparison, esp32-nimble's
BLE_GAP_EVENT_ENC_CHANGE handler silently drops the callback if
ble_gap_conn_find fails (returns BLE_HS_ENOTCONN due to a race in
the NimBLE host stack). This leaves on_authentication_complete
unfired and the pairing state machine stuck, causing the 30s timeout.

Fix with a polling fallback: each bridge poll cycle, query NimBLE
directly via esp_idf_sys::ble_gap_conn_find using the conn_handle
stored in BleState during on_connect. When the link is encrypted
and MTU >= BLE_MTU_MIN, complete the pairing state machine the same
way on_authentication_complete would have.

Key design decisions:
- Gate on confirm_sent_at.is_some() (NC pairing path only) to keep
  Windows Just Works on the original callback path.
- Skip (don't disconnect) when MTU < BLE_MTU_MIN: Android delays ATT
  MTU exchange until after SMP, so MTU 23 is expected immediately
  after encryption and updates to 247 before the GATT write.
- Add pairing_rejected flag so operator rejection prevents a late SMP
  completion (via fallback or on_authentication_complete) from
  promoting a rejected peer to authenticated.
- Add deferred_connected.is_none() and !authenticated guards in
  on_authentication_complete to prevent double Connected events if
  both the fallback and the callback fire.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* modem: address PR 763 review comments

- Fix doc comment on check_encryption_fallback to include pairing_rejected
  in the list of early-exit guard conditions.
- Log unexpected ble_gap_conn_find errors (non-BLE_HS_ENOTCONN) as warnings
  for field diagnostics; silently return only on ENOTCONN.
- Extract complete_pairing_under_lock helper to remove duplicated
  pairing state-machine logic between on_authentication_complete and
  ch... (continued)

15 of 15 new or added lines in 1 file covered. (100.0%)

24453 of 30261 relevant lines covered (80.81%)

203.59 hits per line