Re: [PATCH v3 2/5] fat: allocate persistent inode numbers

["J. Bruce Fields" <bfields@fieldses.org>]

On Mon, Sep 24, 2012 at 09:32:00PM +0900, OGAWA Hirofumi wrote:
> Namjae Jeon <linkinjeon@gmail.com> writes:
> 
> > 2012/9/24, OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>:
> >> Namjae Jeon <linkinjeon@gmail.com> writes:
> >>
> >>>> I see. fileid seems to be stat.ino on nfsd4. inode->i_ino is actually
> >>>> just a hash key of inode hash (exception is only in audit, iirc).
> >>>>
> >>>> So, what happens if we set "stat->ino = i_pos" on fat_getattr().
> >>>>
> >>>> int fat_getattr(struct vfsmount *mnt, struct dentry *dentry, struct
> >>>> kstat
> >>>> *stat)
> >>>> {
> >>>> 	struct inode *inode = dentry->d_inode;
> >>>> 	generic_fillattr(inode, stat);
> >>>> 	stat->blksize = MSDOS_SB(inode->i_sb)->cluster_size;
> >>>> 	if (opts->nfs == FAT_NFS_LIMITED) {
> >>>> 		/* Use i_pos for ino. This is used as fileid of nfs. */
> >>>> 		stat->ino = MSDOS_SB(inode->i_sb)->i_pos;
> >>
> >> 		stat->ino = fat_i_pos_read(MSDOS_SB(inode->i_sb), inode);
> >>
> >> Ouch, I forgot to use fat_i_pos_read().
> >>
> > There is some unclear thing.
> > When I see first mail, I think maybe you don't want to use i_pos for inode->ino.
> > FAT allocate inode->ino from i_unique on server side and If NFS client
> > use i_pos for inode->ino in fat_get_attr, inode numbers on each
> > client/server will still be mismatched.
> >
> > Would you plz give me hint ?
> 
> ->i_ino is long. It can't hold i_pos fully on 32bit arch, so we can't
> use ->i_no to store i_pos, and changing ->i_ino is unnecessary. If
> getattr() returned i_pos as ino, nobody see ->i_ino anymore except
> internal of kernel.

The NFS server must always return the same inode number for the same
filehandle.  To do otherwise is a bug.

> Furthermore I think there is no issue even if server and client didn't
> have same ino. Because client just uses FH (nfs4 seems to be using
> stat.ino though).

The client may expose a different inode number to userspace, but it's
probably the server-provided inode number that it's checking.

(And even if the Linux client didn't currently happen to do that check,
this would still be a bug.)

--b.