Re: [PATCH] fs: report direct io constraints through file_getattr

[Keith Busch <kbusch@kernel.org>]

On Fri, Jul 10, 2026 at 06:35:19AM +0200, Christoph Hellwig wrote:
> On Thu, Jul 09, 2026 at 07:46:42AM -0600, Keith Busch wrote:
> > On Thu, Jul 09, 2026 at 09:13:52AM +0200, Christoph Hellwig wrote:
> > > On Tue, Jul 07, 2026 at 06:18:43PM -0700, Keith Busch wrote:
> > > 
> > > > +	fa->fsx_dio_mem_align = bdev_dma_alignment(bdev) + 1;
> > > > +	fa->fsx_dio_offset_align = bdev_logical_block_size(bdev);
> > > > +	fa->fsx_dio_read_offset_align = bdev_logical_block_size(bdev);
> > > > +	fa->fsx_dio_virt_boundary_align = bdev_virt_boundary_alignment(bdev);
> > > > +	fa->fsx_max_segments = bdev_max_segments(bdev);
> > > 
> > > How is the max_segments value defined in a way that is meaningful to
> > > userspace?
> > 
> > It tells you how many sub-sector vectors you can submit in your
> > readv/writev before it needs to add up to a logical block size.
> > 
> > Ex: 4k logical block size, 4 byte DMA, 256 max segments. You can define
> > 4-byte iov's in your command, but you'll hit the max segment count
> > before you have a valid IO if they're all that small.
> 
> Ah, makes sense.  But besides the missing documentation I think
> max_segments is a bit of a misleading name for that.
> 
> Something like max_vecs_per_block (although we don't expose blocks
> in the UAPI) or max_vecs_per_granularity (I think grammar wants a word
> with me for that, though...) might be a bit more suitable.

The granularity it has to add up to is defined by the
fsx_dio_offset_align attribute. This is a bit long, but to make that
relationship clear, how about:

  fsx_dio_max_vecs_per_offset_align