On Mon, Mar 03, 2025 at 03:22:48PM +0000, Matthew Wilcox wrote: > Funnily, I was looking at the md bitmap code too. It's the last part of > the kernel using buffer heads with not-a-pagecache page, so it's the > only caller of alloc_page_buffers() remaining. > > I think it should just use the page cache to read/write the file data, > but I haven't looked into it in detail.
The md bitmap code is awful as it abuses it's own buffer heads in a way that is non-coherent with the underlying fs. It should just be using the vfs read/write helpers for in-kernel direct I/O with a scoped nofs context.
