Should bio_add_folio() add the folio even it returned false for @len == 0 case?

[Qu Wenruo <wqu@suse.com>]

Hi,

Normally no one should call bio_add_folio() with a zero length, but when 
that happened, bio_add_folio() will always return false.

However if the bio has enough free bvec slots, that folio will still be 
added to that folio, breaking that "returning false meaning addition 
failed" assumption.

This can result folio double freeing (well, if you ignore the bigger bug 
of queuing zero length in the first place), as return false normally 
means the add failed, caller normally needs to release that folio manually.

But caller also needs to cleanup that folio, which involves releasing 
each queued folio, resulting double freeing of the last added folio.

Thus I'm wondering if we should add a @len == 0 check (maybe with a 
WARN_ON_ONCE() too) in bio_add_page().

And please don't ask me why/how I hit this, I'm already embarrassed 
enough to hit it in the first place.

Thanks,
Qu