Re: [PATCH 1/3] md: call del_gendisk in control path

Hi,

在 2026/1/27 10:13, Bart Van Assche 写道:
> On 6/11/25 12:31 AM, Xiao Ni wrote:
>>   static inline int mddev_trylock(struct mddev *mddev)
>>   {
>> -    return mutex_trylock(&mddev->reconfig_mutex);
>> +    int ret;
>> +
>> +    ret = mutex_trylock(&mddev->reconfig_mutex);
>> +    if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
>> +        ret = -ENODEV;
>> +        mutex_unlock(&mddev->reconfig_mutex);
>> +    }
>> +    return ret;
>>   }
>
> This change seems wrong to me. This change will cause mutex_unlock() to
> be called if mutex_trylock() *failed*. Additionally, returning -ENODEV
> from the failure path will cause the caller to call mutex_unlock() a
> second time if mutex_trylock() failed. Please fix this!

oops, this looks correct.

Just wonder why we didn't hit any strange bug with this, a try lock failure
should be quite possible.

>
> As a reminder, from include/linux/mutex.h:
>
> /*
>  * NOTE: mutex_trylock() follows the spin_trylock() convention,
>  *       not the down_trylock() convention!
>  *
>  * Returns 1 if the mutex has been acquired successfully, and 0 on 
> contention.
>  */
>
> Bart.
>
-- 
Thansk,
Kuai