Re: [PATCH v3 2/4] Linux: Add close_range

[Florian Weimer <fweimer@redhat.com>]

* Adhemerval Zanella:

>>> diff --git a/sysdeps/unix/sysv/linux/bits/unistd_ext.h b/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>> index c315cc5cb8..799c59512c 100644
>>> --- a/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>> +++ b/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>> @@ -33,4 +33,15 @@
>>>     not detached and has not been joined.  */
>>>  extern __pid_t gettid (void) __THROW;
>>>  
>>> +/* Unshare the file descriptor table before closing file descriptors.  */I d
>>> +#define CLOSE_RANGE_UNSHARE     (1U << 1)
>> 
>> Needs to be indented.
>> 
>> Please include <linux/close_range.h> if available.  It's a separate
>> header so that we can use it.
>> 
>> I think if you do that, the consistency check won't add much value
>> because it can't really test anything.
>
> I am not very found of adding Linux headers, specially on installed
> headers; but it seems we are already doing it for statx.h. I have 
> added:
>
> --
> #ifdef __has_include
> # if __has_include ("linux/close_range.h")
> #  include "linux/close_range.h"
> # endif
> #endif
> #ifndef CLOSE_RANGE_UNSHARE
> # define CLOSE_RANGE_UNSHARE (1U << 1)
> #endif
> --
>
> This is very recent kernel and it will take some time until we get
> the minimum kernel near the one that starts to provide the close_range
> flags (v5.9).  So I think the consistent check would be useful to
> system that do not use use the latest kernel headers.

My point was that if the kernel headers are too old, there is nothing to
check, and if they contain CLOSE_RANGE_UNSHARE, the built-in definition
is not checked either.  But then the risk of divergence is greatly
reduced with this approach.

Thanks,
Florian