Skip to content

Linux + iPhone Continuity / iMessage / SMS

fix: le reconnect after out-of-range

Zack Bartel48a4d88fb170
diff --git a/docs/BLUETOOTH.md b/docs/BLUETOOTH.md
index 24ff20f..d60028e 100644
--- a/docs/BLUETOOTH.md
+++ b/docs/BLUETOOTH.md
@@ -3940,3 +3940,31 @@ before pairing. Only `0bda:8771` is listed, since it has two reports. CSR clones
no USB ID that tells them apart from working parts.
Not verified: the reporter's `LinkKey Type`.
+
+### 2026-11-04 - ANCS on a Type=5 bond, and no retry after a walk (#303)
+
+| | |
+|---|---|
+| Controller | Mac mini built-in Broadcom BCM20702 (`usb:v05ACp8289`); TP-Link UB500 also tried |
+| BlueZ | 5.87 (Linux Mint 22.3) |
+| Phone | iPhone 16, iOS 26.5 |
+| Tether | 0.3.2 |
+
+The bond's `LinkKey` was `Type=5` (P-192), and `--bt-status` showed `Bearer API: unknown`. No
+`Bearer.LE1` line appears anywhere in the log. Even so, after a re-pair ANCS reached
+`Notification mirroring is active.` That status is only set when the phone answers
+`GetAppAttributes`, so iOS had authorized access. The session dropped at the same moment as the
+`BREDR1` supervision timeout during a walk out of range.
+
+When the phone came back, BR/EDR reconnected and MAP and PBAP worked, but ANCS was never tried
+again. `sync_ancs` kept the device path only while LE read up or within
+`ANCS_BEARER_GRACE_SECONDS` of the drop. LE can never read up without a populated `Bearer.LE1`,
+so once the grace window had passed the path stayed empty. `should_hold_ancs` now also keeps
+the path while the classic link is up and there is no LE bearer, and discovery decides whether
+ANCS answers. The same gap applies on BlueZ without the bearer API (#226).
+
+Not settled: which transport carried ANCS on a bond with no LE keys from CTKD. It could be a
+separate LE pairing that BlueZ never surfaced as `Bearer.LE1`, or ATT over BR/EDR. Apple's ANCS
+spec only describes LE. The bond's `info` sections and a `btmon` capture would settle it. Also
+not verified: whether the phone offers ANCS again after a walk on the fixed build, and that an
+idle phone does not loop on rebuilds.
diff --git a/src/core/include/tether/bluetooth/bearer_supervisor.hpp b/src/core/include/tether/bluetooth/bearer_supervisor.hpp
index 05aa78f..8348848 100644
--- a/src/core/include/tether/bluetooth/bearer_supervisor.hpp
+++ b/src/core/include/tether/bluetooth/bearer_supervisor.hpp
@@ -113,6 +113,14 @@ namespace tether::bluetooth {
return ancs_enabled && bearer.le_available && !bearer.le_dialling && (!bearer.le_connected || link_dead);
}
+ // Whether the ANCS client keeps its device path. Without a populated
+ // Bearer.LE1 the LE state is invisible, so the classic link being up is the
+ // only sign the phone is back. Discovery decides whether ANCS answers.
+ inline bool should_hold_ancs(bool ready, const BearerStatus& bearer, int64_t now, int64_t le_down_since) {
+ return ready || bearer.le_connected || (bearer.classic_connected && !bearer.le_available) ||
+ now - le_down_since < ANCS_BEARER_GRACE_SECONDS;
+ }
+
// Whether the phone has been gone long enough, and for the right reason, to
// lock the session.
inline bool should_lock_on_away(bool enabled,
diff --git a/src/core/src/bluetooth/connection.cpp b/src/core/src/bluetooth/connection.cpp
index b5cf519..33cbd6a 100644
--- a/src/core/src/bluetooth/connection.cpp
+++ b/src/core/src/bluetooth/connection.cpp
@@ -1232,7 +1232,7 @@ namespace tether::bluetooth {
}
// ponytail: one fixed grace window rather than per-phase tuning. Widen it
// if a controller is seen flapping slower than this.
- const bool hold = ready || le_up || now - le_down_since < ANCS_BEARER_GRACE_SECONDS;
+ const bool hold = should_hold_ancs(ready, bearers->status(), now, le_down_since);
std::string device_path;
bool offers_ancs = false;
diff --git a/test/test_bluetooth_supervisor.cpp b/test/test_bluetooth_supervisor.cpp
index acbfb44..461680f 100644
--- a/test/test_bluetooth_supervisor.cpp
+++ b/test/test_bluetooth_supervisor.cpp
@@ -1224,3 +1224,25 @@ TEST(AwayLock, NeverLocksBeforeAnythingHasGoneDown) {
const BearerStatus gone = gone_bearer();
EXPECT_FALSE(should_lock_on_away(true, gone, false, REASON_TIMEOUT, 300, -1, 30));
}
+
+// No Bearer.LE1, so LE never reads up; the phone walked out past the
+// grace window and came back over BR/EDR. ANCS must be tried again.
+TEST(HoldAncs, ClassicReturningHoldsWhenLeIsInvisible) {
+ BearerStatus bearer = gone_bearer();
+ const int64_t past_grace = ANCS_BEARER_GRACE_SECONDS + 60;
+ EXPECT_FALSE(should_hold_ancs(false, bearer, past_grace, 0)) << "phone still away";
+
+ bearer.classic_connected = true;
+ EXPECT_TRUE(should_hold_ancs(false, bearer, past_grace, 0));
+}
+
+// With a populated LE bearer its state is known, so classic alone does not hold.
+TEST(HoldAncs, ClassicAloneDoesNotHoldWhenLeIsVisible) {
+ BearerStatus bearer = gone_bearer();
+ bearer.classic_connected = true;
+ bearer.le_available = true;
+ const int64_t past_grace = ANCS_BEARER_GRACE_SECONDS + 60;
+ EXPECT_FALSE(should_hold_ancs(false, bearer, past_grace, 0));
+ EXPECT_TRUE(should_hold_ancs(false, bearer, ANCS_BEARER_GRACE_SECONDS - 1, 0)) << "inside the grace window";
+ EXPECT_TRUE(should_hold_ancs(true, bearer, past_grace, 0)) << "a live session holds";
+}