Re: [PATCH v7 5/5] linux: Add mount_setattr

[Florian Weimer <fweimer@redhat.com>]

* Adhemerval Zanella:

>> On 11 Jul 2022, at 05:55, Florian Weimer <fweimer@redhat.com> wrote:
>> 
>> * Adhemerval Zanella via Libc-alpha:
>> 
>>> I am not sure how to handle it, glibc sys/mount.h should be a standalone
>>> header.
>> 
>> Is it really necessary to keep <sys/mount.h> standalone?  It's not even
>> a standard interface, and we don't have conform tests for this header as
>> far as I can see.
>> 
>> So I think the usual __has_include approach should work here, too.
>
> The main problem is sys/mount.h defines interfaces for syscall added
> in multiple releases.  It means that depending of the kernel version
> used by the compiler we will need to add some defines (for instance
> fsconfig_command).  There is also a small difference where glibc
> header define the MS_* constant as a unnamed enum (not sure if this
> affects C++ mangling).

You are right, I had missed that.

> I know that using kernel headers simplify some support, specially sync
> with newer releases; but this also adds more complexity where we need
> to handle integration outside glibc project.

For openat2, we got a separate header.  Maybe we can request that for
the fsmount interfaces, too?  Than we can do the usual thing just for
<linux/fsmount.h>.  <linux/mount.h> can still include <linux/fsmount.h>,
so the change would be transparent.

Thanks,
Florian