Re: ext4 regression in v5.9-rc2 from e7bfb5c9bb3d on ro fs with overlapped bitmaps

[Josh Triplett]

On Thu, Oct 08, 2020 at 07:54:09PM -0700, Darrick J. Wong wrote:
> On Thu, Oct 08, 2020 at 03:38:58PM -0700, Josh Triplett wrote:
> > On Thu, Oct 08, 2020 at 10:54:48AM -0700, Darrick J. Wong wrote:
> > > > > the head", and continued with the notion that anything other than
> > > > > e2fsprogs making something to be mounted by mount(2) and handled by
> > > > > fs/ext4 is being "inflicted", and if the goal didn't still seem to be
> > > > > "how do we make it go away so that only e2fsprogs and the kernel ever
> > > > > touch ext4". I started this thread because I'd written some userspace
> > > > > code, a new version of the kernel made that userspace code stop working,
> > > > > so I wanted to report that the moment I'd discovered that, along with a
> > > > > potential way to address it with as little disrupton to ext4 as
> > > > > possible.
> > > 
> > > Ted: I don't think it's a good idea to make SHARED_BLOCKS disable block
> > > validity checking by default.  You could someday enable users to write
> > > to block-sharing files by teaching ext4 how to COW, at which point you'd
> > > need correct bitmaps and block validity to prevent accidental overwrite
> > > of critical metadata.  At that point you'd either be stuck with the
> > > precedent that SHARED_BLOCKS implies noblock_validity, or you'd end up
> > > breaking Josh's images again.
> > 
> > That's a fair point (though I think a writable COW version of
> > SHARED_BLOCKS would need a different flag). It'd make more sense to key
> > this logic off of RO_COMPAT_READONLY or similar. But even better:
> 
> It wouldn't require a new flag -- "rocompat" features bits mean that
> "it's safe to allow userspace to read files off the disk if software
> doesn't recognize this feature bit".

Sure; I was more thinking that if that involved adding some data
structures to track shared blocks and the need to COW, whatever
mechanism that used might potentially need an incompat flag.

> If someone /did/ add the code to write to files safely in the presence
> of shared blocks, all they'd have to do to light up the functionality is
> add it to the _SUPP define.  Also, it's strange that the kernel source
> doesn't even know of SHARED_BLOCKS, but that's also on Ted...

It would be nice if the kernel's ext4.h header and e2fsprogs were fully
in sync, yes.

- Josh Triplett