Comment # 5 on bug 1162471 from Wenruo Qu

(In reply to Dominique Leuenberger from comment #3)
> (In reply to Wenruo Qu from comment #2)
> > A known bug, caused by v5.4 kernel where metadata over-commit behavior
> > changed.
> 
> Caused by 5.4 kernel? How come we only see this with 5.5? (or was that a
> typo?)
> 
> (all in all: glad we had openQA catch it)

It's v5.4. We had reported since v5.4.

We knew it, tried several ways to fix it, but none of them was good enough for
backport, until recent days.

(In reply to Jiri Slaby from comment #4)
> (In reply to Wenruo Qu from comment #2)
> > The fix is already in David's tree, would be sent to upstream in next pull
> > request.
> 
> Could you be more specific? For example a subject or SHA?

Commit 86e7e13f738e ("btrfs: do not zero f_bavail if we have available space")
in David's misc-next branch will fix it.

Although that "Fixes:" tag mentions it's a v4.5 commit, it's really Josef's
metadata space overcommit patches changing the behavior and caused the problem.
(Too many patches in that series, thus I failed to pin to exact commit)

Thanks.