Re: [PATCH 08/10] xfs: move xfs_iomap_write_allocate to xfs_aops.c

[Christoph Hellwig <hch@lst.de>]

On Mon, Feb 11, 2019 at 10:22:45AM -0500, Brian Foster wrote:
> It's an assert in xfs_bmapi_convert_delalloc() FWIW. That also doesn't
> account for the fact that this changes an explicit error into a debug
> mode notification. I'm not familiar with the history of this check and
> the whole xfs_alert_fsblock_zero() thing, but AFAICT it's the only thing
> that prevents an in-core record corruption from constructing a
> superblock overwrite in this path.

Ok, I'll change the earlier patch to return an error.

> Can we please fix this comment to explain what "what the caller expects"
> means? This could be as simple as appending "(i.e., the caller has
> already trimmed against overlapping COW fork blocks)" to the last
> sentence above.

Actually, I think the better idea is to just do the explicit trim
to cow_fsb in the caller, as that is a lot more obvious.

> > +	ASSERT(wpc->imap.br_startoff <= offset_fsb);
> > +	ASSERT(wpc->imap.br_startoff + wpc->imap.br_blockcount >= offset_fsb);
> 
> Looks like this one should be >.

Indeed.