Re: [PATCH v2] thunderbolt: Fix to check return value of ida_simple_get

[Mukesh Ojha <mojha@codeaurora.org>]

On 3/21/2019 11:19 AM, Mika Westerberg wrote:
> On Thu, Mar 21, 2019 at 02:09:41AM +0530, Mukesh Ojha wrote:
>> On 3/20/2019 9:59 PM, Mika Westerberg wrote:
>>> On Wed, Mar 20, 2019 at 11:24:45AM -0500, Aditya Pakki wrote:
>>>> In enumerate_services, ida_simple_get on failure can return an error and
>>>> leaks memory during device_register failure. The patch ensures that
>>>> the dev_set_name is set on non failure cases, and releases memory in
>>>> case of failure.
>>>>
>>>> Signed-off-by: Aditya Pakki <pakki001@umn.edu>
>>>>
>>>> ---
>>>> v1: Missed cleanup of svc in case of allocation failure and
>>>> device_register failure.
>>>> ---
>>>>    drivers/thunderbolt/xdomain.c | 9 ++++++++-
>>>>    1 file changed, 8 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/thunderbolt/xdomain.c b/drivers/thunderbolt/xdomain.c
>>>> index e27dd8beb94b..eb08275185bf 100644
>>>> --- a/drivers/thunderbolt/xdomain.c
>>>> +++ b/drivers/thunderbolt/xdomain.c
>>>> @@ -740,6 +740,7 @@ static void enumerate_services(struct tb_xdomain *xd)
>>>>    	struct tb_service *svc;
>>>>    	struct tb_property *p;
>>>>    	struct device *dev;
>>>> +	int id;
>>>>    	/*
>>>>    	 * First remove all services that are not available anymore in
>>>> @@ -768,7 +769,12 @@ static void enumerate_services(struct tb_xdomain *xd)
>>>>    			break;
>>>>    		}
>>>> -		svc->id = ida_simple_get(&xd->service_ids, 0, 0, GFP_KERNEL);
>>>> +		id = ida_simple_get(&xd->service_ids, 0, 0, GFP_KERNEL);
>>>> +		if (id < 0) {
>>>> +			kfree(svc);
>>>> +			break;
>>>> +		}
>>>> +		svc->id = id;
>>>>    		svc->dev.bus = &tb_bus_type;
>>>>    		svc->dev.type = &tb_service_type;
>>>>    		svc->dev.parent = &xd->dev;
>>>> @@ -776,6 +782,7 @@ static void enumerate_services(struct tb_xdomain *xd)
>>>>    		if (device_register(&svc->dev)) {
>>>>    			put_device(&svc->dev);
>>>> +			kfree(svc);
>>> You can't do this after device_register() is called. The put_device()
>>> above is sufficient.
>>
>> If  device_register fails, how would svc gets freed? we need  to kfree svc
>> here as well.
> Please read the comment on top of device_register(). It should explain.
>
> So no kfree here.

Thanks for pointer Mika.

Overlooked the fact that dev is a data member of svc not a pointer, also 
noticed there are many places in tree where , it is not
followed.

Thanks,
Mukesh