On Mon, May 30, 2022 at 08:57:15AM -0400, Mikulas Patocka wrote: > In the kernel 5.17 and before, if the "invalidatepage" method is NULL, the > kernel will use block_invalidatepage (see do_invalidatepage). So, we don't > have to provide explicit "invalidatepage" in 5.17 and before and we don't > have to backport this bugfix there. > > Note that the commit 7ba13abbd31e contains this piece of code: > -#ifdef CONFIG_BLOCK > - if (!invalidatepage) > - invalidatepage = block_invalidatepage; > -#endif > > So, it explicitly breaks filesystems that have NULL invalidatepage and > NULL invalidate_folio and that relied on block_invalidatepage being called > implicitly. So, I believe this commit is the root cause of this bug. Oh, right, I missed ntfs3 in that commit. Oops. Acked-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>
