Re: [PATCH] bus: fsl-mc: Fix the double free in fsl_mc_device_add()

[Su Hui <suhui@nfschina.com>]

On 2024/11/14 16:57, Dan Carpenter wrote:
> On Thu, Nov 14, 2024 at 11:41:25AM +0300, Dan Carpenter wrote:
>> On Thu, Nov 14, 2024 at 04:27:52PM +0800, Su Hui wrote:
>>> Clang static checker(scan-build) warning：
>>> drivers/bus/fsl-mc/fsl-mc-bus.c: line 909, column 2
>>> Attempt to free released memory.
>>>
>>> When 'obj_desc->type' == "dprc" and begin to free 'mc_bus' and 'mc_dev',
>>> there is a double free problem because of 'mc_dev = &mc_bus->mc_dev'.
>>> Add a judgment to fix this problem.
>>>
>>> Fixes: a042fbed0290 ("staging: fsl-mc: simplify couple of deallocations")
>>> Signed-off-by: Su Hui <suhui@nfschina.com>
>>> ---
>>>   drivers/bus/fsl-mc/fsl-mc-bus.c | 6 ++++--
>>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
>>> index 930d8a3ba722..8d2d5d3cc782 100644
>>> --- a/drivers/bus/fsl-mc/fsl-mc-bus.c
>>> +++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
>>> @@ -905,8 +905,10 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
>>>   
>>>   error_cleanup_dev:
>>>   	kfree(mc_dev->regions);
>>> -	kfree(mc_bus);
>>> -	kfree(mc_dev);
>>> +	if (strcmp(mc_dev->obj_desc.type, "dprc") == 0)
>> This works, but it would probably be nicer to write this as:
>>
>> 	if (is_fsl_mc_bus_dprc(mc_dev))
>> 		kfree(mc_bus);
>> 	else
>> 		kfree(mc_dev);
>>
>> That way it would match the release function.
Yes, it's better!
>     820          mc_dev->dev.release = fsl_mc_device_release;
>
> 	[ snip ]
>
>     891           * The device-specific probe callback will get invoked by device_add()
>     892           */
>     893          error = device_add(&mc_dev->dev);
>     894          if (error < 0) {
>     895                  dev_err(parent_dev,
>     896                          "device_add() failed for device %s: %d\n",
>     897                          dev_name(&mc_dev->dev), error);
>     898                  goto error_cleanup_dev;
>
> I don't think this goto is correct.  I think fsl_mc_device_release() will be
> called automaticall on this path so the goto is a double free.
Agreed too, maybe using put_device(&mc_dev->dev) to replace ?
>
>     899          }
>     900
>     901          dev_dbg(parent_dev, "added %s\n", dev_name(&mc_dev->dev));
>     902
>     903          *new_mc_dev = mc_dev;
>     904          return 0;
>     905
>     906  error_cleanup_dev:
>     907          kfree(mc_dev->regions);
>     908          if (is_fsl_mc_bus_dprc(mc_dev))

Yep, this looks better :)

I will send a v2 patch to use is_fsl_mc_bus_dprc().
It might take some time because I'm not sure about using
'put_device(&mc_dev->dev)' to replace 'goto error_cleanup_dev'.

regards,
Su Hui