Re: Regression in handling power cuts since 3a1e819b4e80 ("ovl: store file handle of lower inode on copy up")

[Rafał Miłecki <zajec5@xxxxxxxxx>]

On Sat, 20 Oct 2018 at 08:58, Richard Weinberger <richard@xxxxxx> wrote:
> Am Freitag, 19. Oktober 2018, 23:28:33 CEST schrieb Rafał Miłecki:
> > On Fri, 19 Oct 2018 at 16:59, Richard Weinberger <richard@xxxxxx> wrote:
> > > Am Freitag, 19. Oktober 2018, 16:45:53 CEST schrieb Richard Weinberger:
> > > > Well, I fear it uncovers a problem in UBIFS. We had already problems with overlayfs.
> > > > Did you bisect the problem and you are sure that the said commit is the first bad commit?
> > >
> > > Another thing, UBIFS has no export operations, so overlayfs will fall back to xattrs,
> > > if I read the commit correctly.
> > > Maybe this is the root of the problem.
> > >
> > > I'm currently traveling, so I cannot test much. Please make sure whether
> > > 3a1e819b4e80 is really the first bad commit.
> > > Then I can go for a bug hunt. :-)
> >
> > I did another test and everything keeps pointing to the 3a1e819b4e80.
> >
> > 1) I switched to the 4.12.14
> > 2) I confirmed problem exists there
> > 3) I added "return 0;" at the beginning of ovl_set_origin()
> > 4) I confirmed problem disappeared
>
> Okay. But I still think we're facing some UBIFS problem.
>
> To double check, on Linux 4.12.14 the problem exists and by reverting 3a1e819b4e80 (or making it a nop)
> it goes away?

Yes. This is exactly what I described above (4 steps test). Making
ovl_set_origin() a NOP in 4.12.14 makes problem disappear.

> Do you see the warning regarding the double orphan only in the bad case?

No. Let me quote myself: "Please note I wrote this error appears
*with* ovl commit and also with one commit earlier.". If I checkout
3a1e819b4e80~1 (one commit before) I don't see power cut corruption
BUT I still see orphan error.


It seems despite me explaining some things twice still not everything
is clear. Let me work on a complete summary for all commits &
behaviors.

-- 
Rafał

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/