[PATCH v2] erofs-utils: mkfs: use scandir for stable output

[Jooyung Han]

Hi Gao,

Since the new patch is quite different from the last, I sent it as a
new one, instead of up-revving.


On Tue, Dec 3, 2024 at 3:58 PM Gao Xiang <hsiangkao at linux.alibaba.com> wrote:
>
>
>
> On 2024/12/3 14:37, Jooyung Han wrote:
> > (I forgot to reply all)
>
> Yeah, I received that.
>
> >
> > On Tue, Dec 3, 2024 at 11:36 AM Gao Xiang <hsiangkao at linux.alibaba.com> wrote:
> >>
> >>
> >>
> >> On 2024/12/3 10:20, Jooyung Han wrote:
> >>> Hi Gao,
> >>>
> >>> I found that in the loop erofs_iget_from_srcpath() is called in
> >>> different order due to readdir and erofs_iget_from_srcpath() calls
> >>> erofs_new_inode() which fills i_ino[0] for newly created inode. I
> >>> think this i_ino[0] having different values caused the difference in
> >>> the output.
> >>
> >> Oh, okay, that makes sense, I think we'd better move
> >>
> >> inode->i_ino[0] = sbi->inos++;  /* inode serial number */
> >>
> >> to erofs_mkfs_dump_tree()  (since we'd better to leave
> >> i_ino[0] stable even without dumping from localdir later.)
> >> and even clean up a bit.
> >>
> >> If you don't have more time to clean up this, let's just
> >> commit a patch to fix this directly.
> >
> > Sounds good to me ;-)
> > To be honest, I don't know the stuff well enough for this cleanup.
>
> Ok, anyway if you could submit a patch that moves
> inode->i_ino[0] = sbi->inos++;
> from erofs_new_inode() to erofs_prepare_inode_buffer()
>
> I will apply directly.
>
> Also I'd like to update it as
> inode->i_ino[0] = ++sbi->inos;
>
> Since `i_ino == 0` is invalid...
>
> Thanks,
> Gao Xiang
>
> >
> >>
> >> Thanks,
> >> Gao Xiang
> >
> > Thanks,
> > Jooyung
>