Re: [PATCH v2 1/5] mips: Do not malloc on getdents64 fallback

[Adhemerval Zanella <adhemerval dot zanella at linaro dot org>]

On 28/08/2019 18:23, Florian Weimer wrote:
> * Adhemerval Zanella:
> 
>> On 28/08/2019 11:42, Florian Weimer wrote:
>>> * Adhemerval Zanella:
>>>
>>>>    struct kernel_dirent
>>>> +  {
>>>> +    unsigned long d_ino;
>>>> +    unsigned long d_off;
>>>> +    unsigned short int d_reclen;
>>>> +    char d_name[1];
>>>> +  } kbuf[KBUF_SIZE / sizeof (struct kernel_dirent)];
>>>
>>> I think it's still not clear to me in which cases we actually need to
>>> move the dirent entries in the buffer.  My impression is that we just
>>> need to move d_name by one byte because before, d_type was after the
>>> name, and afterwards, it comes before the name.  But the record
>>> boundaries are unchanged.
>>
>> My understanding is the record boundary would be same as long the 
>> d_name fits on the alignment padding space minus the size of the
>> d_type.  Otherwise the dirent64 will need to be extended.
> 
> Hmm.  The problem is mips64 n32, right?  Where unsigned long is 32 bits?

Yes, it would require either a mips64-n32 specific path or implementation.