Re: [PATCH v3] xfs: add FALLOC_FL_WRITE_ZEROES to XFS code base

[Christoph Hellwig <hch@infradead.org>]

On Wed, Oct 29, 2025 at 11:22:55AM -0700, Darrick J. Wong wrote:
> > +	if (mode & FALLOC_FL_WRITE_ZEROES) {
> > +		if (xfs_is_cow_inode(ip) || !bdev_write_zeroes_unmap_sectors(
> 
> xfs_is_cow_inode() only tells us if the inode is capable of doing out of
> place writes.  Why would a regular reflinked inode be ineligible for
> WRITE_ZEROES?

Yes, this shoyuld be xfs_is_always_cow_inode.

> I don't understand why this bdev_write_zeroes_unmap_sectors check is
> here and not in xfs_alloc_file_space.  Shouldn't other callers of
> xfs_alloc_file_space be restricted from passing in XFS_BMAPI_ZERO if the
> block device doesn't support unmap_sectors?

Othere callers are fine with the software fallback for the block zeroing
helpers.  But this is a good question that should probably be documented
in a comment.