fix(ble): bound the write, and correct two reply-path edge cases

Follow-up to review of the two preceding commits.

The write lock added in 00135bb prevented concurrent writes from dropping the
link, but bounded nothing. CommandHandler's own timeout does not cover the
write: send() awaits _sender_func() and only afterwards arms
asyncio.wait(futures, timeout=...). So a stalled write -- observed on hardware
running to minutes -- held the lock indefinitely while every other command
queued behind it, with nothing logged, no error raised and no DISCONNECTED
event. Before the lock only the stalled command hung; the others went out and
could trip the error-19 disconnect, which at least recovered. The lock turned
a bounded, self-healing failure into an unbounded silent one, and also blocked
the post-reconnect CMD_APP_START behind the dead connection's holder.

The write (lock acquisition included) is now bounded by
BLEConnection.WRITE_TIMEOUT. On expiry the link is torn down rather than the
lock merely released: the underlying CoreBluetooth write may still be in
flight, and a second write racing it re-creates the overlap the lock exists to
prevent. Tearing down hands over to the reconnect path, which is bounded and
self-healing.

The lock is now a lazily-created property, mirroring _mesh_request_lock in
commands/base.py, so an instance built without __init__ still works. The two
BLE tests previously assigned _write_lock themselves, which meant deleting the
__init__ line left them green; there is now a test that __init__ provides it.

Also in send_anon_req:

- A failed change_contact_path() no longer proceeds. The device still has the
  contact as flood, so sendAnonReq() floods the request, and the server gates
  REGIONS/OWNER/BASIC behind isRouteDirect() and drops it -- the caller then
  waits out a full path-scaled timeout for a reply that cannot arrive. It now
  returns ERROR path_reset_failed.
- encode_reply_path() clamps to the server's 64-byte reply_path buffer, which
  MyMesh.cpp memcpys into with no length check, and rejects hash mode 3 (the
  4-byte hops Packet::isValidPathLen refuses). Not a regression -- the previous
  encoder overflowed identically -- but this function is the chokepoint and its
  comment claimed to bound the read.
- The hop count saturates at 63 instead of being masked with & 63, which would
  wrap a 64-hop path to zero hops, i.e. request a zero-hop reply from a distant
  node. Not reachable from a device-sourced contact (the reader caps the field
  at 63) but silent if it ever were.
- The suggested_timeout multiplier now scales by the hops actually emitted
  rather than the contact's claimed out_path_len, which can differ once the
  encoder clamps or truncates.

Correction to 30446ed's message: the claim that mode 0 is "byte-for-byte
identical" is wrong. Differentially, over 30000 randomised contact fields
restricted to what a device can actually emit, mode 0 diverges in 1177 of
10118 cases -- every one of them a path containing a 0x00 byte, and in every
one the old encoder was the wrong one. The accurate claim is "unchanged for
mode-0 paths containing no 0x00 byte".

src/meshcore/ble_cx.py

@@ -28,6 +28,11 @@ UART_RX_CHAR_UUID = "6E400002-B5A3-F393-E0A9-E50E24DCCA9E"
 UART_TX_CHAR_UUID = "6E400003-B5A3-F393-E0A9-E50E24DCCA9E"
 
 class BLEConnection:
+    # Upper bound on a single write (lock acquisition included). Healthy writes
+    # measured 0.06-0.17s against real hardware; observed stalls ran 20s to
+    # minutes, so this preempts them rather than waiting for CoreBluetooth.
+    WRITE_TIMEOUT = 10.0
+
     def __init__(self, address=None, device=None, client=None, pin=None):
         """
         Constructor: specify address or an existing BleakClient.
@@ -53,13 +58,27 @@ class BLEConnection:
         self.rx_char = None
         self._disconnect_callback = None
         self._background_tasks: set[asyncio.Task] = set()
-        # Serialises write_gatt_char(). Two overlapping writes to the same
+        self._write_lock_obj: Optional[asyncio.Lock] = None
-        # characteristic drop the link outright (observed on macOS/CoreBluetooth:
+
-        # "BLE write failed: 19", connection gone). Nothing above this layer
+    @property
-        # guarantees callers are sequential -- schedulers, health checks and
+    def _write_lock(self) -> asyncio.Lock:
-        # user commands all issue commands independently -- so the transport has
+        """Serialises write_gatt_char().
-        # to enforce it. Lazily created so it binds to the running loop.
+
-        self._write_lock: Optional[asyncio.Lock] = None
+        Two overlapping writes to the same characteristic drop the link outright
+        (observed on macOS/CoreBluetooth: "BLE write failed: 19", connection
+        gone). Nothing above this layer guarantees callers are sequential --
+        schedulers, health checks and user commands all issue independently --
+        so the transport has to enforce it.
+
+        Lazily created so it binds to the running loop, mirroring the
+        _mesh_request_lock property in commands/base.py. Read through getattr so
+        an instance built without __init__ still works.
+        """
+        lock = getattr(self, "_write_lock_obj", None)
+        if lock is None:
+            lock = asyncio.Lock()
+            self._write_lock_obj = lock
+        return lock
 
     def _spawn_background(self, coro) -> asyncio.Task:
         """Create a tracked background task (prevents GC of fire-and-forget tasks)."""
@@ -198,6 +217,10 @@ class BLEConnection:
         if self.reader is not None:
             self._spawn_background(self.reader.handle_rx(data))
 
+    async def _write_locked(self, data):
+        async with self._write_lock:
+            await self.client.write_gatt_char(self.rx_char, bytes(data), response=True)
+
     async def send(self, data):
         if not self.client:
             logger.error("Client is not connected")
@@ -207,11 +230,24 @@ class BLEConnection:
         if not self.rx_char:
             logger.error("RX characteristic not found")
             return False
-        if self._write_lock is None:
+        # Bound the whole acquire-plus-write. A stalled write has been seen to
-            self._write_lock = asyncio.Lock()
+        # hang for minutes, and CommandHandler's own timeout does not cover this
+        # -- it starts only after _sender_func returns -- so without a bound the
+        # serialising lock would queue every other command behind the stall
+        # indefinitely, with nothing logged and no disconnect raised. Turning one
+        # hung command into a silent whole-client stall would be worse than the
+        # overlap the lock exists to prevent.
         try:
-            async with self._write_lock:
+            await asyncio.wait_for(self._write_locked(data), timeout=self.WRITE_TIMEOUT)
-                await self.client.write_gatt_char(self.rx_char, bytes(data), response=True)
+        except asyncio.TimeoutError:
+            # Do not simply release and carry on: the underlying write may still
+            # be in flight, and a second write racing it re-creates the exact
+            # overlap that kills the link. Tear the connection down so the
+            # reconnect path takes over -- bounded and self-healing.
+            logger.warning(f"BLE write timed out after {self.WRITE_TIMEOUT}s")
+            if self._disconnect_callback:
+                await self._disconnect_callback("ble_write_timeout")
+            return False
         except Exception as exc:
             logger.warning(f"BLE write failed: {exc}")
             if self._disconnect_callback:

src/meshcore/commands/base.py

@@ -57,6 +57,13 @@ def _validate_destination(dst: DestinationType, prefix_length: int = 6) -> bytes
         )
 
 
+# Size of the server-side reply_path buffer (uint8_t reply_path[64] in
+# simple_repeater/MyMesh.h). It is memcpy'd into without a length check.
+MAX_REPLY_PATH_BYTES = 64
+# reply_path_len is the low 6 bits of the header byte.
+MAX_REPLY_PATH_HOPS = 63
+
+
 def encode_reply_path(out_path_len: int, out_path_hex: str, out_path_hash_mode: int) -> bytes:
     """Encode the reply path a server should use when answering us.
 
@@ -71,17 +78,36 @@ def encode_reply_path(out_path_len: int, out_path_hex: str, out_path_hash_mode:
 
     The path itself is reversed by *hop*, not by byte: a return path visits the
     same hops in the opposite order, and each hop's multi-byte hash must stay
-    intact. (At hash mode 0 the two are indistinguishable, which is why this
+    intact. (For single-byte hops the two are indistinguishable, which is most
-    went unnoticed - mode 0 is the default.)
+    of why this went unnoticed - mode 0 is the default.)
     """
     hash_mode = max(out_path_hash_mode, 0)      # -1 means "flood", i.e. no path
+    if hash_mode > 2:
+        # The server computes hash_size = mode + 1, and Packet::isValidPathLen
+        # rejects 4-byte hops outright, so such a path is unusable on the wire.
+        logger.warning(
+            f"Unsupported out_path_hash_mode {out_path_hash_mode}; "
+            "requesting a zero-hop reply path instead"
+        )
+        return b"\x00"
     hash_size = hash_mode + 1
-    hops = max(out_path_len, 0) & 63
+    # Saturate rather than mask: `& 63` would silently wrap a 64-hop path to
+    # zero hops, i.e. a zero-hop reply for a distant node.
+    hops = min(max(out_path_len, 0), MAX_REPLY_PATH_HOPS)
 
     raw = bytes.fromhex(out_path_hex or "")
     # Never read past what the contact actually carries; a truncated or padded
     # field would otherwise yield short trailing hops.
     hops = min(hops, len(raw) // hash_size)
+    # The server memcpys into a fixed 64-byte reply_path with no bounds check
+    # (simple_repeater/MyMesh.cpp), so never describe more than fits.
+    max_hops = min(MAX_REPLY_PATH_HOPS, MAX_REPLY_PATH_BYTES // hash_size)
+    if hops > max_hops:
+        logger.warning(
+            f"Reply path of {hops} hops x {hash_size}B exceeds the "
+            f"{MAX_REPLY_PATH_BYTES}B the server can hold; truncating to {max_hops}"
+        )
+        hops = max_hops
 
     path = b"".join(
         raw[i * hash_size:(i + 1) * hash_size] for i in range(hops - 1, -1, -1)
@@ -362,7 +388,15 @@ class CommandHandlerBase:
             if contact["out_path_len"] == -1:
                 logger.info("No path set trying zero hop")
                 zero_hop = True
-                await self.change_contact_path(contact, "")
+                path_res = await self.change_contact_path(contact, "")
+                if path_res is not None and path_res.type == EventType.ERROR:
+                    # The device still has this contact as flood, so sendAnonReq
+                    # will flood the request -- and the server gates REGIONS,
+                    # OWNER and BASIC behind isRouteDirect(), silently dropping
+                    # it. Better to fail here than to wait out a full timeout
+                    # for a reply that cannot come.
+                    logger.error("Could not set zero-hop path, aborting anon request")
+                    return Event(EventType.ERROR, {"reason": "path_reset_failed"})
             # update_contact() normally reflects the change back onto the dict, so
             # out_path_len reads 0 here. Clamp anyway: if that call failed (e.g. the
             # device query inside it errored) the dict is still -1, and the unsigned
@@ -388,7 +422,8 @@ class CommandHandlerBase:
 
             exp_tag = result.payload["expected_ack"].hex()
             # Use provided timeout or fallback to suggested timeout (with 5s default)
-            result.payload["suggested_timeout"] = result.payload.get("suggested_timeout", 4000) * (out_path_len + 1) # update timeout from path_len
+            emitted_hops = reply_path[0] & 63   # what actually went on the wire
+            result.payload["suggested_timeout"] = result.payload.get("suggested_timeout", 4000) * (emitted_hops + 1) # update timeout from path_len
             actual_timeout = timeout if timeout is not None and timeout > 0 else result.payload.get("suggested_timeout", 4000) / 800.0
             actual_timeout = min_timeout if actual_timeout < min_timeout else actual_timeout
             self._reader.register_binary_request(pubkey_prefix.hex(), exp_tag, request_type, actual_timeout, context=context, is_anon=True)

tests/unit/test_error_handling.py

@@ -368,34 +368,30 @@ async def test_send_anon_req_flood_contact_still_forces_and_restores_zero_hop(
     assert sent == b"\x39" + bytes.fromhex(VALID_PUBKEY_HEX) + b"\x01" + b"\x00"
 
 
-async def test_send_anon_req_survives_change_contact_path_not_mutating(
+async def test_send_anon_req_aborts_when_change_contact_path_fails(
     command_handler, mock_connection, mock_dispatcher
 ):
-    """A failed change_contact_path must not crash and must still restore the path.
+    """A failed zero-hop switch must abort, not send an unanswerable request.
 
-    update_contact() normally writes out_path_len back onto the dict, but if it
+    If change_contact_path() fails the device still has the contact as flood, so
-    errors the value stays -1. Unclamped, the unsigned to_bytes raises
+    sendAnonReq() floods the request -- and the server gates REGIONS/OWNER/BASIC
-    OverflowError, which skips reset_path and leaves the contact pinned to
+    behind isRouteDirect(), dropping it silently. Sending anyway would burn a
-    zero-hop on the device.
+    full path-scaled timeout waiting for a reply that cannot arrive.
     """
     contact = {"public_key": VALID_PUBKEY_HEX, "out_path_len": -1, "out_path": ""}
     command_handler._get_contact_by_prefix = MagicMock(return_value=contact)
-    # Returns an error without reflecting the change onto the dict.
     command_handler.change_contact_path = AsyncMock(
         return_value=Event(EventType.ERROR, {"reason": "device_query_failed"})
     )
     command_handler.reset_path = AsyncMock()
-    setup_event_response(
-        mock_dispatcher, EventType.MSG_SENT,
-        {"expected_ack": b"\x01\x02\x03\x04", "suggested_timeout": 4000},
-    )
 
     result = await command_handler.send_anon_req(VALID_PUBKEY_HEX, MagicMock(value=1))
 
-    assert not result.is_error()
+    assert result.is_error()
-    sent = mock_connection.send.await_args.args[0]
+    assert result.payload["reason"] == "path_reset_failed"
-    assert sent == b"\x39" + bytes.fromhex(VALID_PUBKEY_HEX) + b"\x01" + b"\x00"
+    mock_connection.send.assert_not_awaited()
-    command_handler.reset_path.assert_awaited_once()
+    # Nothing was changed on the device, so there is nothing to restore.
+    command_handler.reset_path.assert_not_awaited()
 
 
 # ── send_trace handles unknown path_hash_len without NameError ──
@@ -427,7 +423,6 @@ async def test_ble_send_serialises_concurrent_writes():
 
     conn = BLEConnection.__new__(BLEConnection)   # bypass bleak availability check
     conn._disconnect_callback = None
-    conn._write_lock = None
     conn.rx_char = object()
 
     overlap = {"current": 0, "max": 0}
@@ -451,7 +446,6 @@ async def test_ble_send_releases_lock_on_write_failure():
     from meshcore.ble_cx import BLEConnection
 
     conn = BLEConnection.__new__(BLEConnection)
-    conn._write_lock = None
     conn.rx_char = object()
     reasons = []
 
@@ -595,3 +589,128 @@ async def test_reader_contact_out_path_keeps_zero_bytes():
     assert contact["out_path_len"] == 1
     assert contact["out_path_hash_mode"] == 2
     assert contact["out_path"] == "aa00bb", "the 0x00 inside the hop hash was lost"
+
+
+# ── BLE write bound (a stalled write must not wedge the client) ──
+
+async def test_ble_write_timeout_tears_down_the_link():
+    """A stalled write must be bounded and must drop the link, not just release.
+
+    CommandHandler's timeout starts only after the write returns, so nothing
+    else bounds this. Releasing the lock alone would be wrong too: the
+    underlying write may still be in flight, and a second write racing it
+    re-creates the overlap the lock exists to prevent.
+    """
+    from meshcore.ble_cx import BLEConnection
+
+    conn = BLEConnection.__new__(BLEConnection)
+    conn.rx_char = object()
+    reasons = []
+
+    async def capture(reason):
+        reasons.append(reason)
+
+    conn._disconnect_callback = capture
+    conn.WRITE_TIMEOUT = 0.05
+
+    class Stalling:
+        async def write_gatt_char(self, char, data, response=True):
+            await asyncio.Event().wait()      # never completes
+
+    conn.client = Stalling()
+
+    result = await asyncio.wait_for(conn.send(b"\x01"), timeout=2.0)
+    assert result is False
+    assert reasons == ["ble_write_timeout"]
+
+
+async def test_ble_stalled_write_does_not_block_later_commands_forever():
+    """Head-of-line: queued writes must not inherit an unbounded stall."""
+    from meshcore.ble_cx import BLEConnection
+
+    conn = BLEConnection.__new__(BLEConnection)
+    conn.rx_char = object()
+    conn._disconnect_callback = None
+    conn.WRITE_TIMEOUT = 0.05
+
+    class Stalling:
+        async def write_gatt_char(self, char, data, response=True):
+            await asyncio.Event().wait()
+
+    conn.client = Stalling()
+
+    # Five writers behind one stalled holder must all return, not hang.
+    done = await asyncio.wait_for(
+        asyncio.gather(*(conn.send(bytes([i])) for i in range(5))), timeout=3.0
+    )
+    assert done == [False] * 5
+
+
+async def test_ble_connection_init_provides_the_lock():
+    """__init__ must set up the lock slot; the tests must not paper over it."""
+    from meshcore.ble_cx import BLEConnection
+
+    conn = BLEConnection.__new__(BLEConnection)
+    BLEConnection.__init__(conn, address="AA:BB:CC:DD:EE:FF")
+    assert conn._write_lock_obj is None          # lazily created
+    assert isinstance(conn._write_lock, asyncio.Lock)
+    assert conn._write_lock is conn._write_lock  # stable across reads
+
+
+# ── reply-path bounds ────────────────────────────────────────
+
+async def test_encode_reply_path_rejects_unsupported_hash_mode():
+    # mode 3 -> 4-byte hops, which Packet::isValidPathLen refuses outright.
+    assert encode_reply_path(2, "aabbccddeeffaabb", 3) == b"\x00"
+
+
+async def test_encode_reply_path_never_overflows_the_server_buffer():
+    """The server memcpys into a fixed 64-byte reply_path with no length check."""
+    for mode in (0, 1, 2):
+        hash_size = mode + 1
+        out = encode_reply_path(63, "ab" * 200, mode)
+        body = out[1:]
+        assert len(body) <= 64, f"mode {mode} emitted {len(body)} bytes"
+        assert (out[0] & 63) * hash_size == len(body), "header disagrees with body"
+
+
+async def test_send_anon_req_timeout_tracks_emitted_hops_not_claimed_length():
+    """Scaling must follow the hops actually sent, not a length that got clamped."""
+    command_handler = CommandHandler()
+    command_handler.dispatcher = MagicMock()
+    command_handler._reader = MagicMock()
+    conn = MagicMock()
+    conn.send = AsyncMock()
+
+    async def sender(data):
+        await conn.send(data)
+
+    command_handler._sender_func = sender
+    # Claims 8 hops but carries only 3 bytes -> 3 hops actually emitted.
+    command_handler._get_contact_by_prefix = MagicMock(return_value={
+        "public_key": VALID_PUBKEY_HEX,
+        "out_path_len": 8,
+        "out_path": "aabbcc",
+        "out_path_hash_mode": 0,
+    })
+    setup_event_response(
+        command_handler.dispatcher, EventType.MSG_SENT,
+        {"expected_ack": b"\x01\x02\x03\x04", "suggested_timeout": 1000},
+    )
+
+    result = await command_handler.send_anon_req(VALID_PUBKEY_HEX, MagicMock(value=1))
+
+    sent = conn.send.await_args.args[0]
+    assert sent[-4] & 63 == 3                     # 3 hops on the wire
+    assert result.payload["suggested_timeout"] == 4000   # (3 + 1) x 1000, not 9x
+
+
+async def test_encode_reply_path_saturates_rather_than_wrapping():
+    """hops is a 6-bit field; masking would wrap 64 to 0.
+
+    A 64-hop path would then be described as zero hops -- a zero-hop reply
+    request for a distant node -- instead of being clamped and logged.
+    """
+    out = encode_reply_path(64, "ab" * 64, 0)
+    assert out[0] & 63 == 63
+    assert len(out[1:]) == 63