4.9-stable review patch. If anyone has any objections, please let me know.
------------------ From: Christoph Hellwig <[email protected]> commit 54a4ef8af4e0dc5c983d17fcb9cf5fd25666d94e upstream. We currently fall back from direct to buffered writes if we detect a remaining shared extent in the iomap_begin callback. But by the time iomap_begin is called for the potentially unaligned end block we might have already written most of the data to disk, which we'd now write again using buffered I/O. To avoid this reject all writes to reflinked files before starting I/O so that we are guaranteed to only write the data once. The alternative would be to unshare the unaligned start and/or end block before doing the I/O. I think that's doable, and will actually be required to support reflinks on DAX file system. But it will take a little more time and I'd rather get rid of the double write ASAP. Signed-off-by: Christoph Hellwig <[email protected]> Reviewed-by: Brian Foster <[email protected]> Reviewed-by: Darrick J. Wong <[email protected]> Signed-off-by: Darrick J. Wong <[email protected]> [slight changes in context due to the new direct I/O code in 4.10+] Signed-off-by: Greg Kroah-Hartman <[email protected]> --- fs/xfs/xfs_aops.c | 45 --------------------------------------------- fs/xfs/xfs_file.c | 9 +++++++++ fs/xfs/xfs_trace.h | 2 +- 3 files changed, 10 insertions(+), 46 deletions(-) --- a/fs/xfs/xfs_aops.c +++ b/fs/xfs/xfs_aops.c @@ -1263,44 +1263,6 @@ xfs_map_trim_size( bh_result->b_size = mapping_size; } -/* Bounce unaligned directio writes to the page cache. */ -static int -xfs_bounce_unaligned_dio_write( - struct xfs_inode *ip, - xfs_fileoff_t offset_fsb, - struct xfs_bmbt_irec *imap) -{ - struct xfs_bmbt_irec irec; - xfs_fileoff_t delta; - bool shared; - bool x; - int error; - - irec = *imap; - if (offset_fsb > irec.br_startoff) { - delta = offset_fsb - irec.br_startoff; - irec.br_blockcount -= delta; - irec.br_startblock += delta; - irec.br_startoff = offset_fsb; - } - error = xfs_reflink_trim_around_shared(ip, &irec, &shared, &x); - if (error) - return error; - - /* - * We're here because we're trying to do a directio write to a - * region that isn't aligned to a filesystem block. If any part - * of the extent is shared, fall back to buffered mode to handle - * the RMW. This is done by returning -EREMCHG ("remote addr - * changed"), which is caught further up the call stack. - */ - if (shared) { - trace_xfs_reflink_bounce_dio_write(ip, imap); - return -EREMCHG; - } - return 0; -} - STATIC int __xfs_get_blocks( struct inode *inode, @@ -1438,13 +1400,6 @@ __xfs_get_blocks( if (imap.br_startblock != HOLESTARTBLOCK && imap.br_startblock != DELAYSTARTBLOCK && (create || !ISUNWRITTEN(&imap))) { - if (create && direct && !is_cow) { - error = xfs_bounce_unaligned_dio_write(ip, offset_fsb, - &imap); - if (error) - return error; - } - xfs_map_buffer(inode, bh_result, &imap, offset); if (ISUNWRITTEN(&imap)) set_buffer_unwritten(bh_result); --- a/fs/xfs/xfs_file.c +++ b/fs/xfs/xfs_file.c @@ -554,6 +554,15 @@ xfs_file_dio_aio_write( if ((iocb->ki_pos & mp->m_blockmask) || ((iocb->ki_pos + count) & mp->m_blockmask)) { unaligned_io = 1; + + /* + * We can't properly handle unaligned direct I/O to reflink + * files yet, as we can't unshare a partial block. + */ + if (xfs_is_reflink_inode(ip)) { + trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); + return -EREMCHG; + } iolock = XFS_IOLOCK_EXCL; } else { iolock = XFS_IOLOCK_SHARED; --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -3353,7 +3353,7 @@ DEFINE_INODE_IREC_EVENT(xfs_reflink_conv DEFINE_RW_EVENT(xfs_reflink_reserve_cow); DEFINE_RW_EVENT(xfs_reflink_allocate_cow_range); -DEFINE_INODE_IREC_EVENT(xfs_reflink_bounce_dio_write); +DEFINE_SIMPLE_IO_EVENT(xfs_reflink_bounce_dio_write); DEFINE_IOMAP_EVENT(xfs_reflink_find_cow_mapping); DEFINE_INODE_IREC_EVENT(xfs_reflink_trim_irec);
