Re: [syzbot] [iomap?] WARNING in ifs_free

[Christoph Hellwig]

On Thu, Feb 19, 2026 at 04:46:58PM -0800, Joanne Koong wrote:
> The folio is uptodate but the ifs uptodate bitmap is not reflected as
> fully uptodate. I think this is because ntfs3 handles writes for
> compressed files through its own interface that doesn't go through
> iomap where it calls folio_mark_uptodate() but the ifs bitmap doesn't
> get updated. fuse-blk servers that operate in writethrough mode run
> into something like this as well [2].
> 
> This doesn't lead to any data corruption issues. Should we get rid of
> the  WARN_ON_ONCE(ifs_is_fully_uptodate(folio, ifs) !=
> folio_test_uptodate(folio))? The alternative is to make a modified
> version of the functionality in "iomap_set_range_uptodate()" a public
> api callable by subsystems.

I'd honestly rather have ntfs3 come along and explain what their
doing.  They've copy and pasted large chunks of the buffered
read code for now reason, which already annoys me and I'd rather
not paper over random misuses.