Re: [RFC v3 1/5] block: add block layer APIs resembling Linux ZonedBlockDevice ioctls.

[Damien Le Moal]

On 6/29/22 10:50, Sam Li wrote:
>>>>>>> +    rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct 
>>>>>>> blk_zone);
>>>>>>> +    g_autofree struct blk_zone_report *rep = g_new(struct 
>>>>>>> blk_zone_report, nrz);
>>>>>>
>>>>>> g_new() looks incorrect. There should be 1 struct blk_zone_report
>>>>>> followed by nrz struct blk_zone structs. Please use g_malloc(rep_size)
>>>>>> instead.
>>>>>
>>>>> Yes! However, it still has a memory leak error when using g_autofree
>>>>> && g_malloc.
>>>>
>>>> That may be because you are changing the value of the rep pointer while
>>>> parsing the report ?
>>>
>>> I am not sure it is the case. Can you show me some way to find the problem?
>>
>> Not sure. I never used this g_malloc()/g_autofree() before so not sure how
>> it works. It may be that g_autofree() work only with g_new() ?
>> Could you try separating the declaration and allocation ? e.g.
>>
>> Declare at the beginning of the function:
>> g_autofree struct blk_zone_report *rep = NULL;
>>
>> And then when needed do:
>>
>> rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone);
>> rep = g_malloc(rep_size);
> 
> Actually, the memory leak occurs in that way. When using zone_mgmt,
> memory leak still occurs. Asan gives the error information not much so
> I haven't tracked down the problem yet.

See this:

https://blog.fishsoup.net/2015/11/05/attributecleanup-mixed-declarations-and-code-and-goto/

Maybe you can find some hints.

-- 
Damien Le Moal
Western Digital Research