Re: RFC: MFC'ng a change to struct mount

On Fri, May 19, 2023 at 08:00:02AM -0700, Rick Macklem wrote:
> Hi,
> 
> Since there seemed to be interest in it, I have MFC'd most of
> the changes needed to run nfsd(8) in a vnet jail to stable/13.
> 
> However, there is one change that *might* affect the VFS KAPI.
> 
> I need to add a pointer called mnt_exjail to struct mount.
> 
> There is a field called mnt_pad0. It is 32bits, but is followed
> by a field that is a structure.  As such, replacing it with the pointer
> does not change the offsets of subsequent fields for amd64.
> --> Since mnt_exjail is only used in generic code, this should
>      not break the VFS KAPI for amd64, I think?
> I will check i386, but since it will be replacing a 32bit field with a
> 32bit pointer, I suspect it will be ok.
> 
> Does anyone think some other architecture will be broken by this
> or know an easy way I can check all arches?

Replacing mnt_pad0 with a pointer should be ok.  It's proceeded by a
pointer and thus properly aligned and followed by struct statfs which
contains uint64_t's so there is indeed a gap after mnt_pad0 on 64-bit
systems.

-- Brooks