Re: [PATCH RFC] iomap: ensure iomap_dio_bio_iter() only submit bios that are fs block aligned

[Matthew Wilcox <willy@infradead.org>]

On Mon, Oct 06, 2025 at 09:05:15PM -0700, Christoph Hellwig wrote:
> On Tue, Oct 07, 2025 at 04:52:09AM +0100, Matthew Wilcox wrote:
> > On Tue, Oct 07, 2025 at 01:40:22PM +1030, Qu Wenruo wrote:
> > > During my development of btrfs bs > ps support, I hit some read bios
> > > that are submitted from iomap to btrfs, but are not aligned to fs block
> > > size.
> > > 
> > > In my case the fs block size is 8K, the page size is 4K. The ASSERT()
> > > looks like this:
> > 
> > Why isn't bdev_logical_block_size() set correctly by btrfs?
> 
> bdev_logical_block_size is never set by the file system.  It is the LBA
> size of the underlying block device.  But if the file system block size
> is larger AND the file system needs to do file system block size
> granularity operations that is not the correct boundary.  See also the
> iov_iter_alignment for always COW / zoned file system in
> xfs_file_dio_write.

But the case he's complaining about is bs>PS, so the LBA size really is
larger than PAGE_SIZE.