On Thu, Jan 11, 2024 at 07:57:08PM +0000, Matthew Wilcox wrote: > I'm just doing an xfstests run now, but I think this will fix the problem > (it's up to 400 seconds after crashing at 6 seconds the first time). > Essentially if we pass in a length larger than the folio we find, we > just clamp the amount of work we do to the size of the folio. No, that's not enough. It dies in submit_bh_wbc() at the BUG_ON(buffer_delay(bh)); I have no idea why. Since Dave tells me he's not using generic_perform_write(), and the only possible user would be the bs>PS work, I'm going to abandon this for now.
