Re: [RFC][PATCH v3 01/10] fs: common implementation of file type

[Phillip Potter <phil@philpotter.co.uk>]

On Wed, Oct 24, 2018 at 05:41:15PM +0300, Amir Goldstein wrote:
> On Wed, Oct 24, 2018 at 4:02 PM Theodore Y. Ts'o <tytso@mit.edu> wrote:
> >
> > On Tue, Oct 23, 2018 at 09:19:53PM +0100, Phillip Potter wrote:
> > > diff --git a/include/linux/file_type.h b/include/linux/file_type.h
> >
> > Shouldn't this be in include/uapi/linux/fs_types.h?
> >
> 
> IDGI. Why do we want this file in uapi?
> The DT_ constants are already defined by glibc dirent.h
> and the FT_ constants and macros we don't want to expose
> to uapi at all. Right?
> 
> Maybe all we need is a comment above DT_ constants
> that those are defined by POSIX and in glibc dirent.h?
> 
> > One of things which must be made crystal clear is these definitions
> > MUST NOT ever change.  It would break the Userspace ABI, and would
> > break file systems on-disk format.
> >
> > It might also be useful to be clear *why* we are making this change in
> > the first place.  Code refactorization is good from a code maintenance
> > perspective (either to fix bugs, although this code is pretty
> > trivial),
> 
> Very trivial code that has had an out of bounds access bug for two
> decades and bug was duplicated to 7 filesystems. IMO, fixing the bug in
> one place instead of 7 is a good enough reason for re-factoring.
> 
> Thanks,
> Amir.

Dear Amir,

Regarding the location of the extra header file, I'm happy to stick it
in either include/linux or include/uapi/linux before sending out the
revised series. You make a valid point about the FT_ constants and macros
being kernel only though, so perhaps the comment is a better option? I
will wait a bit before sending out as I'm interested to see what Ted thinks.
Many thanks.

Regards,
Phil