For I/O to reflinked blocks we always need to write an entire new file system block, and the code enforces the file system block alignment for the entire file if it has any reflinked blocks. Use the new STATX_DIO_READ_ALIGN flag to report the asymmetric read vs write alignments for reflinked files. Signed-off-by: Christoph Hellwig <hch@xxxxxx> --- fs/xfs/xfs_iops.c | 24 +++++++++++++++++++++--- 1 file changed, 21 insertions(+), 3 deletions(-) diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index 6b0228a21617..053d05f5567d 100644 --- a/fs/xfs/xfs_iops.c +++ b/fs/xfs/xfs_iops.c @@ -580,9 +580,27 @@ xfs_report_dioalign( struct xfs_buftarg *target = xfs_inode_buftarg(ip); struct block_device *bdev = target->bt_bdev; - stat->result_mask |= STATX_DIOALIGN; + stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN; stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; - stat->dio_offset_align = bdev_logical_block_size(bdev); + stat->dio_read_offset_align = bdev_logical_block_size(bdev); + + /* + * On COW inodes we are forced to always rewrite an entire file system + * block or RT extent. + * + * Because applications assume they can do sector sized direct writes + * on XFS we fall back to buffered I/O for sub-block direct I/O in that + * case. Because that needs to copy the entire block into the buffer + * cache it is highly inefficient and can easily lead to page cache + * invalidation races. + * + * Tell applications to avoid this case by reporting the natively + * supported direct I/O read alignment. + */ + if (xfs_is_cow_inode(ip)) + stat->dio_offset_align = xfs_inode_alloc_unitsize(ip); + else + stat->dio_offset_align = stat->dio_read_offset_align; } static void @@ -658,7 +676,7 @@ xfs_vn_getattr( stat->rdev = inode->i_rdev; break; case S_IFREG: - if (request_mask & STATX_DIOALIGN) + if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) xfs_report_dioalign(ip, stat); if (request_mask & STATX_WRITE_ATOMIC) xfs_report_atomic_write(ip, stat); -- 2.45.2
