[PATCH] Bluetooth: fix autoconnect for pending connect attempt

[PATCH] Bluetooth: fix autoconnect for pending connect attempt

[Jakub Pawlowski]

When adding device to auto connect whitelist when there is pending
connect attempt, there is no need to update scan, or to add it to
pend_le_conns list.

When trying to connect to le device, it is added to pend_le_conns and
background scan is updated. There's no need to repeat this operation when
adding device to auto connect list. Only update of params->auto_connect
value is required.

If both operations try to update background scan, and are quickly queued
together when scan was disabled, second operation might improperly try to
start, instead of restarting scan. This means that adding device to
connect whitelist would report failure, even though it succeeded.

In order to reproduce this bug type in bluetoothctl:
connect D9:00:00:00:00
disconnect D9:00:00:00:00
connect D9:00:00:00:00

and observe bluetoothd logs (error happens during second connect attempt):
src/device.c:device_connect_le() Connection attempt to: D0:5F:B8:52:22:9F
Failed to add device D0:5F:B8:52:22:9F (1): Busy (0x0a)
---
 net/bluetooth/mgmt.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/net/bluetooth/mgmt.c b/net/bluetooth/mgmt.c
index ccaf5a4..1108281 100644
--- a/net/bluetooth/mgmt.c
+++ b/net/bluetooth/mgmt.c
@@ -6122,7 +6122,12 @@ static int hci_conn_params_set(struct hci_request *req, bdaddr_t *addr,
 		break;
 	case HCI_AUTO_CONN_DIRECT:
 	case HCI_AUTO_CONN_ALWAYS:
-		if (!is_connected(hdev, addr, addr_type)) {
+		/* If we are connected, don't add to whitelist. If we are
+		 * connecting, we're already added to pend_le_conns and
+		 * scanning.
+		 */
+		if (!is_connected(hdev, addr, addr_type) &&
+		    !params->auto_connect == HCI_AUTO_CONN_EXPLICIT) {
 			list_add(&params->action, &hdev->pend_le_conns);
 			__hci_update_background_scan(req);
 		}
-- 
2.6.0.rc2.230.g3dd15c0

Re: [PATCH] Bluetooth: fix autoconnect for pending connect attempt

[Marcel Holtmann]

Hi Jakub,

> When adding device to auto connect whitelist when there is pending
> connect attempt, there is no need to update scan, or to add it to
> pend_le_conns list.
> 
> When trying to connect to le device, it is added to pend_le_conns and
> background scan is updated. There's no need to repeat this operation when
> adding device to auto connect list. Only update of params->auto_connect
> value is required.
> 
> If both operations try to update background scan, and are quickly queued
> together when scan was disabled, second operation might improperly try to
> start, instead of restarting scan. This means that adding device to
> connect whitelist would report failure, even though it succeeded.
> 
> In order to reproduce this bug type in bluetoothctl:
> connect D9:00:00:00:00
> disconnect D9:00:00:00:00
> connect D9:00:00:00:00
> 
> and observe bluetoothd logs (error happens during second connect attempt):
> src/device.c:device_connect_le() Connection attempt to: D0:5F:B8:52:22:9F
> Failed to add device D0:5F:B8:52:22:9F (1): Busy (0x0a)
> ---
> net/bluetooth/mgmt.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/net/bluetooth/mgmt.c b/net/bluetooth/mgmt.c
> index ccaf5a4..1108281 100644
> --- a/net/bluetooth/mgmt.c
> +++ b/net/bluetooth/mgmt.c
> @@ -6122,7 +6122,12 @@ static int hci_conn_params_set(struct hci_request *req, bdaddr_t *addr,
> 		break;
> 	case HCI_AUTO_CONN_DIRECT:
> 	case HCI_AUTO_CONN_ALWAYS:
> -		if (!is_connected(hdev, addr, addr_type)) {
> +		/* If we are connected, don't add to whitelist. If we are
> +		 * connecting, we're already added to pend_le_conns and
> +		 * scanning.
> +		 */
> +		if (!is_connected(hdev, addr, addr_type) &&
> +		    !params->auto_connect == HCI_AUTO_CONN_EXPLICIT) {

we are not doing it like this. I think you are looking for != here.

Regards

Marcel