Re: [PATCH] linux: Use stat_overflow to check overflow in fstatat

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 29/01/2021 19:25, Stafford Horne wrote:
> On Wed, Jan 27, 2021 at 10:28:58AM -0300, Adhemerval Zanella wrote:
>>
>>
>> On 27/01/2021 09:40, Stafford Horne wrote:
>>> Ping Adhemerval on this one.
>>>
>>> I still thing the below is needed for the STAT_IS_KERNEL_STAT
>>> cases.
>>>
>>
>> Why OpenRISC is not setting XSTAT_IS_XSTAT64 ? As a possible newer
>> ABI, I see no pointing in support old non-LFS ABI.
> 
> I got it all working without this patch after:
> 
>   - Setting XSTAT_IS_XSTAT64 and a few other kernel_stats.h
>   - Updating my shlib-version to 2.33
> 
> Note to myself in glibc LFS stands for Large File System. I see that in many
> parts of the stats code but couldn't find the actual acronym explained, I found
> it in the manual, it was hard to search for on the internet.
> 
> Thanks.
> 
> -Stafford
> 

I think I will need to revise the internal stat defaults, the whole 
idea of my stat consolidation is to avoid this very of issue for 
newer ports.

Ideally newer ports should not have to support non-LFS calls, which
means that sysdeps/unix/sysv/linux/[f,l]stat[at].c should be empty
TU.

However it seems that the default XSTAT_IS_XSTAT64 is still 0 on
both the Linux and generic one (the generic defines it to 1 for
__WORDSIZE == 64 at least).

For 2.34 I will consolidate the kernel_stat.h, so the default values
will be:

  #define STAT_IS_KERNEL_STAT 1
  #define XSTAT_IS_XSTAT64    1
  #define STATFS_IS_STATFS64  1

The STAT_IS_KERNEL_STAT is only used for non-LFS call and old xstat
support so it wouldn't matter for newer ABIs.  The only important flag
here is XSTAT_IS_XSTAT64 (which is currently misleading since xstat
is the compat symbols now).

It would probably need to consolidate the statfs code, which is another
messy one.