[PATCH] posix: Fix generic p{read,write}v buffer allocation (BZ#22457)

[Florian Weimer]

* Adhemerval Zanella:

> On 23/11/2017 17:08, Florian Weimer wrote:
>> * Adhemerval Zanella:
>> 
>>> +  size_t mmap_size = ALIGN_UP (bytes, GLRO(dl_pagesize));
>>> +  void *buffer = __mmap (NULL, mmap_size, PROT_READ | PROT_WRITE,
>>> +		         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>>> +  if (__glibc_unlikely (buffer == MAP_FAILED)
>>> +      || __glibc_unlikely (!PTR_IS_ALIGNED (buffer, GLRO(dl_pagesize))))
>>>      return -1;
>> 
>> I don't think POSIX requires that the length of the mapping is a
>> multiple of the page size.  You could probably drop the alignment
>> check, too, because I don't really see any reason why the alignment
>> requirement would be related to the page size.

> Indeed we do not need to align to page size, however the alignment
> requirement was the motivation to actually change its implementation
> (check commit message for c79a72aa5cb8357c216a71015c7448a9259c8531).

I meant that there is no way to query the alignment required for
O_DIRECT, and that page size alignment might be insufficient.  Some
file systems hard-code a 4K alignment requirement, and microblaze has
at least 4K pages, so we should be good.  But it feels strange to
write a check against the wrong constant.