Re: linux-next: manual merge of the vfs-brauner tree with the bcachefs tree

[Amir Goldstein <amir73il@gmail.com>]

On Thu, Oct 26, 2023 at 9:35 PM Kent Overstreet
<kent.overstreet@linux.dev> wrote:
>
> On Thu, Oct 26, 2023 at 08:16:14AM +0300, Amir Goldstein wrote:
> > On Thu, Oct 26, 2023 at 2:02 AM Stephen Rothwell <sfr@canb.auug.org.au> wrote:
> > >
> > > Hi all,
> > >
> > > Today's linux-next merge of the vfs-brauner tree got a conflict in:
> > >
> > >   include/linux/exportfs.h
> > >
> > > between commit:
> > >
> > >   85e95ca7cc48 ("bcachefs: Update export_operations for snapshots")
> > >
> > > from the bcachefs tree and commit:
> > >
> > >   2560fa66d2ac ("exportfs: define FILEID_INO64_GEN* file handle types")
> > >
> > > from the vfs-brauner tree.
> > >
> > > I fixed it up (see below) and can carry the fix as necessary. This
> > > is now fixed as far as linux-next is concerned, but any non trivial
> > > conflicts should be mentioned to your upstream maintainer when your tree
> > > is submitted for merging.  You may also want to consider cooperating
> > > with the maintainer of the conflicting tree to minimise any particularly
> > > complex conflicts.
> > >
> > > --
> > > Cheers,
> > > Stephen Rothwell
> >
> > [adding exportfs maintainers]
> >
> > >
> > > diff --cc include/linux/exportfs.h
> > > index be9900cc8786,21bae8bfeef1..000000000000
> > > --- a/include/linux/exportfs.h
> > > +++ b/include/linux/exportfs.h
> > > @@@ -98,12 -98,17 +98,23 @@@ enum fid_type
> > >          */
> > >         FILEID_FAT_WITH_PARENT = 0x72,
> > >
> > >  +      /*
> > >  +       * 64 bit inode number, 32 bit subvolume, 32 bit generation number:
> > >  +       */
> > >  +      FILEID_BCACHEFS_WITHOUT_PARENT = 0x80,
> > >  +      FILEID_BCACHEFS_WITH_PARENT = 0x81,
> > >  +
> > > +       /*
> > > +        * 64 bit inode number, 32 bit generation number.
> > > +        */
> > >  -      FILEID_INO64_GEN = 0x81,
> > > ++      FILEID_INO64_GEN = 0x82,
> > > +
> > > +       /*
> > > +        * 64 bit inode number, 32 bit generation number,
> > > +        * 64 bit parent inode number, 32 bit parent generation.
> > > +        */
> > >  -      FILEID_INO64_GEN_PARENT = 0x82,
> > > ++      FILEID_INO64_GEN_PARENT = 0x83,
> > > +
> >
> > This is wrong.
> > Those are filesystem defined constants.
> > Please don't change them.
> >
> > 0x81/0x82 have been used by xfs and fuse for years,
> > even though neither defined a constant in this enum so far.
>
> Perhaps we could get that fixed...?

commit 2560fa66d2ac ("exportfs: define FILEID_INO64_GEN*
file handle types") fixes that for fuse.
I may fix up xfs to use these constants later.

>
> > Conflicting with FILEID_BCACHEFS_WITH_PARENT is not
> > a serious issue, but I encourage Kent to pick different constants
> > for bcachefs or keep the bcachefs constants out of this enum.
>
> Happy to do so. Since it seems this enum doesn't have all the constants
> I'd need to avoid conflicting with, I might need some help here :)
>

Technically, you don't *need* to avoid conflicting with fileid types
of other filesystems and you do not *need* to define your constant
in this enum. It serves no real purpose unless your constant
declares a fileid format that other filesystems also use.

See the comment at the top of the enum.

> > It is a slight inconvenience for users that have bcachefs exported
> > to NFS clients and upgrade their server, but maybe that is acceptable.
> > In overlayfs, we encoded type OVL_FILEID_V0 and switched to encoding
> > type OVL_FILEID_V1, but we still accept decoding of both types, neither
> > of which are listed in this enum BTW.
> >
> > Adding fid types to this enum is not required.
> > This enum is a place to standardize and for different fs to share the same
> > fid type/encoding as is the case with  FILEID_INO{32,64}_GEN*.
> > IMO, the bcachefs constant do not follow the convention in this
> > enum and their format is unlikely to be used by other fs, so
> > they should not be added to this enum at all.
>
> Eh?
>
> Most of the constants here appear to be completely filesystem specific -
> I see UDF, nilfs, btrfs, fat...
>

There is no good reason for those to be in the enum either
other than documentation.

> And since you also don't want conflicts with fid_types that aren't
> defined here, it seems like they really should all be here.

If you define your constants internally in bcachefs, I don't care
about conflicts, but if I were you, I would avoid conflicts with
the known types.

If you want to define your constants in this enum please choose
any vacant 0x?{1,2} values. 0xb{1,2}?

Thanks,
Amir.