[RFC PATCH] Bluetooth: core: Allow bind HCI socket user channel when HCI is UP.

[RFC PATCH] Bluetooth: core: Allow bind HCI socket user channel when HCI is UP.

[Vasyl Vavrychuk]

This is needed for user-space to ensure that HCI init scheduled from
hci_register_dev is completed.

Function hci_register_dev queues power_on workqueue which will run
hci_power_on > hci_dev_do_open. Function hci_dev_do_open sets HCI_INIT
for some time.

It is not allowed to bind to HCI socket user channel when HCI_INIT is
set. As result, bind might fail when user-space program is run early
enough during boot.

Now, user-space program can first issue HCIDEVUP ioctl to ensure HCI
init scheduled at hci_register_dev was completed.

Signed-off-by: Vasyl Vavrychuk <vasyl.vavrychuk@opensynergy.com>
---
 net/bluetooth/hci_sock.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index 33b3c0ffc339..c98de809f856 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -1194,9 +1194,7 @@ static int hci_sock_bind(struct socket *sock, struct sockaddr *addr,
 
 		if (test_bit(HCI_INIT, &hdev->flags) ||
 		    hci_dev_test_flag(hdev, HCI_SETUP) ||
-		    hci_dev_test_flag(hdev, HCI_CONFIG) ||
-		    (!hci_dev_test_flag(hdev, HCI_AUTO_OFF) &&
-		     test_bit(HCI_UP, &hdev->flags))) {
+		    hci_dev_test_flag(hdev, HCI_CONFIG)) {
 			err = -EBUSY;
 			hci_dev_put(hdev);
 			goto done;
-- 
2.30.2

Re: [RFC PATCH] Bluetooth: core: Allow bind HCI socket user channel when HCI is UP.

[Marcel Holtmann]

Hi Vasyl,

> This is needed for user-space to ensure that HCI init scheduled from
> hci_register_dev is completed.
> 
> Function hci_register_dev queues power_on workqueue which will run
> hci_power_on > hci_dev_do_open. Function hci_dev_do_open sets HCI_INIT
> for some time.
> 
> It is not allowed to bind to HCI socket user channel when HCI_INIT is
> set. As result, bind might fail when user-space program is run early
> enough during boot.
> 
> Now, user-space program can first issue HCIDEVUP ioctl to ensure HCI
> init scheduled at hci_register_dev was completed.
> 
> Signed-off-by: Vasyl Vavrychuk <vasyl.vavrychuk@opensynergy.com>
> ---
> net/bluetooth/hci_sock.c | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
> index 33b3c0ffc339..c98de809f856 100644
> --- a/net/bluetooth/hci_sock.c
> +++ b/net/bluetooth/hci_sock.c
> @@ -1194,9 +1194,7 @@ static int hci_sock_bind(struct socket *sock, struct sockaddr *addr,
> 
> 		if (test_bit(HCI_INIT, &hdev->flags) ||
> 		    hci_dev_test_flag(hdev, HCI_SETUP) ||
> -		    hci_dev_test_flag(hdev, HCI_CONFIG) ||
> -		    (!hci_dev_test_flag(hdev, HCI_AUTO_OFF) &&
> -		     test_bit(HCI_UP, &hdev->flags))) {
> +		    hci_dev_test_flag(hdev, HCI_CONFIG)) {
> 			err = -EBUSY;
> 			hci_dev_put(hdev);
> 			goto done;

I am not following the reasoning here. It is true that the device has to run init before you can do something with it. From mgmt interface your device will only be announced when it is really ready.

Regards

Marcel

Re: [RFC PATCH] Bluetooth: core: Allow bind HCI socket user channel when HCI is UP.

[Vasyl Vavrychuk]

Hi, Marcel,

On 4/22/2022 12:20 PM, Marcel Holtmann wrote:
> Hi Vasyl,
> 
>> This is needed for user-space to ensure that HCI init scheduled from
>> hci_register_dev is completed.
>>
>> Function hci_register_dev queues power_on workqueue which will run
>> hci_power_on > hci_dev_do_open. Function hci_dev_do_open sets HCI_INIT
>> for some time.
>>
>> It is not allowed to bind to HCI socket user channel when HCI_INIT is
>> set. As result, bind might fail when user-space program is run early
>> enough during boot.
>>
>> Now, user-space program can first issue HCIDEVUP ioctl to ensure HCI
>> init scheduled at hci_register_dev was completed.
>>
>> Signed-off-by: Vasyl Vavrychuk <vasyl.vavrychuk@opensynergy.com>
>> ---
>> net/bluetooth/hci_sock.c | 4 +---
>> 1 file changed, 1 insertion(+), 3 deletions(-)
>>
>> diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
>> index 33b3c0ffc339..c98de809f856 100644
>> --- a/net/bluetooth/hci_sock.c
>> +++ b/net/bluetooth/hci_sock.c
>> @@ -1194,9 +1194,7 @@ static int hci_sock_bind(struct socket *sock, struct sockaddr *addr,
>>
>> 		if (test_bit(HCI_INIT, &hdev->flags) ||
>> 		    hci_dev_test_flag(hdev, HCI_SETUP) ||
>> -		    hci_dev_test_flag(hdev, HCI_CONFIG) ||
>> -		    (!hci_dev_test_flag(hdev, HCI_AUTO_OFF) &&
>> -		     test_bit(HCI_UP, &hdev->flags))) {
>> +		    hci_dev_test_flag(hdev, HCI_CONFIG)) {
>> 			err = -EBUSY;
>> 			hci_dev_put(hdev);
>> 			goto done;
> 
> I am not following the reasoning here. It is true that the device has to run init before you can do something with it. From mgmt interface your device will only be announced when it is really ready.

Sorry, I am not familiar with mgmt interface. I obtain device using 
HCIGETDEVLIST.

BTW. I have pushed related patch [1]. Comparing to this patch, [1] is 
less intrusive since it does not effect user-space semantics.

Patch [1] allows to ensure that device is not in HCI_INIT state by running

     hciconfig hci0 down

This will either wait for HCI_INIT complete and then powers HCI down, or 
cancels pending power_on.

If we apply [1], we can still consider an optimization to allow binding 
during HCI_INIT since this optimization will allow me to ommit extra

     hciconfig hci0 down

[1]: 
https://lore.kernel.org/linux-bluetooth/20220426081823.21557-1-vasyl.vavrychuk@opensynergy.com/T/#u

Kind regards,
Vasyl