nRF5 SDK is not maintained anymore
More Info: Consider nRF Connect SDK for new designs

When are peer managers connection handles invalidated?

Hi,

I have a central that uses PM to connect to some peripherals.

If I connect to a peer (connection handle 0), and disconnect this device (e.g. by shutting it off). I receive BLE_GAP_EVT_DISCONNECTED and all works fine.

But later when I call

pm_conn_handle_get(peerId, &connHandle);

the connection handle I get is still 0 (zero). But it is not a valid connection handle, since e.g. sd_ble_gap_disconnect() fails with BLE_ERROR_INVALID_CONN_HANDLE. I would expect that om_conn_handle_get() returns me BLE_CONN_HANDLE_INVALID.

Although if I connect to the device 1 (connection handle 0), then disconnect it and connect device 2 (connection handle 0, too because it is reused), pm_conn_handle_get() returns BLE_CONN_HANDLE_INVALID.

In my case this leads to a issue where I think the peer is still connected because the connection handle is valid, but in fact it is not and sd_ble_gap_disconnect() returns BLE_ERROR_INVALID_CONN_HANDLE. Sure, I could ignore this error code, but then something else could go wrong without me noticing it.

I there a way to update the peer manager on disconnects?

Best regards

  • I found some more:

    in ble_conn_state.c:

    typedef struct
    {
        nrf_atflags_t valid_flags;                                 /**< Flags indicating which connection handles are valid. */
        nrf_atflags_t connected_flags;                             /**< Flags indicating which connections are connected, since disconnected connection handles will not immediately be invalidated. */
        nrf_atflags_t central_flags;                               /**< Flags indicating in which connections the local device is the central. */
        nrf_atflags_t encrypted_flags;                             /**< Flags indicating which connections are encrypted. */
        nrf_atflags_t mitm_protected_flags;                        /**< Flags indicating which connections have encryption with protection from man-in-the-middle attacks. */
        nrf_atflags_t lesc_flags;                                  /**< Flags indicating which connections have bonded using LE Secure Connections (LESC). */
        nrf_atflags_t user_flags[BLE_CONN_STATE_USER_FLAG_COUNT];  /**< Flags that can be reserved by the user. The flags will be cleared when a connection is invalidated, otherwise, the user is wholly responsible for the flag states. */
    } ble_conn_state_flag_collections_t;

    static void record_set_disconnected(uint16_t conn_handle)
    {
        nrf_atflags_clear(&m_bcs.flags.connected_flags, conn_handle);
    }

    bool ble_conn_state_valid(uint16_t conn_handle)
    {
        if (conn_handle >= BLE_CONN_STATE_MAX_CONNECTIONS)
        {
            return false;
        }
        return nrf_atflags_get(&m_bcs.flags.valid_flags, conn_handle);
    }

    ble_conn_state_valid() is used in pm_conn_handle_get() and checks the valid_flags.. But record_set_disconnected() clears the connected flag not the valid_flags.

    Also the comment for connected_flags says that connection handles will no immediately be invalidated.

    So maybe one should check the connected flag in pm_conn_handle_get()?

  • mgux said:
    I use this snippet to disconnect

    Seems like you are using both connectionHandle and connHandle?

    Try this instead:

    // disconnect and remove peer
    uint16_t connHandle = BLE_CONN_HANDLE_INVALID;
    auto ret = pm_conn_handle_get(peerId, &connHandle);
    APP_ERROR_CHECK(ret);
    
    if(connHandle == BLE_CONN_HANDLE_INVALID)
    {
        return;
    }
    ret = sd_ble_gap_disconnect(connHandle, BLE_HCI_REMOTE_USER_TERMINATED_CONNECTION);
    
    APP_ERROR_CHECK(ret);
    // The peer got removed from the list of bonds.
    ret = pm_peer_delete(peerId);
    APP_ERROR_CHECK(ret);

  • Sorry, that was a copy&paste error.

    I use connHandle everywhere.

    I tried to change im__conn_handle_get to the following:

    uint16_t im_conn_handle_get(pm_peer_id_t peer_id)
    {
        if (peer_id == PM_PEER_ID_INVALID)
        {
            return BLE_CONN_HANDLE_INVALID;
        }
    
        for (uint16_t conn_handle = 0; conn_handle < IM_MAX_CONN_HANDLES; conn_handle++)
        {
            if ((m_connections[conn_handle].peer_id == peer_id) && ble_conn_state_status(conn_handle) == BLE_CONN_STATUS_CONNECTED)
            {
                return conn_handle;
            }
        }
        return BLE_CONN_HANDLE_INVALID;
    }

    ble_conn_state_status() checks the valid_flags and connected_flags. This fixes my problem.

  • mgux said:
    ble_conn_state_status() checks the valid_flags and connected_flags. This fixes my problem.

     Yeah, I see now that disconnected links are first invalidated when a device connects again (record_purge_disconnected() called on BLE_GAP_EVT_CONNECTED in ble_conn_state.c). I haven't looked into why it's implemented like this yet, but an alternative to your patch, could be to call ble_conn_state_status(connHandle) in your application code, and based on the return value of this, then do a disconnect and delete the bond info. With this approach, you don't need to modify the id_manager.c file

Related