On Wed, Oct 30, 2024 at 09:27:41PM +0530, Ritesh Harjani (IBM) wrote: > atomic writes is currently only supported for single fsblock and only > for direct-io. We should not return -ENOTBLK for atomic writes since we > want the atomic write request to either complete fully or fail > otherwise. We should not fallback to buffered-io in case of DIO atomic > write requests. > Let's also catch if this ever happens by adding some WARN_ON_ONCE before > buffered-io handling for direct-io atomic writes. > > More details of the discussion [1]. > > [1]: https://lore.kernel.org/linux-xfs/cover.1729825985.git.ritesh.list@xxxxxxxxx/T/#m9dbecc11bed713ed0d7a486432c56b105b555f04 > > Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@xxxxxxxxx> > --- > fs/ext4/file.c | 7 +++++++ > fs/ext4/inode.c | 14 +++++++++----- > 2 files changed, 16 insertions(+), 5 deletions(-) > > diff --git a/fs/ext4/file.c b/fs/ext4/file.c > index 8116bd78910b..61787a37e9d4 100644 > --- a/fs/ext4/file.c > +++ b/fs/ext4/file.c > @@ -599,6 +599,13 @@ static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) > ssize_t err; > loff_t endbyte; > > + /* > + * There is no support for atomic writes on buffered-io yet, > + * we should never fallback to buffered-io for DIO atomic > + * writes. > + */ > + WARN_ON_ONCE(iocb->ki_flags & IOCB_ATOMIC); > + > offset = iocb->ki_pos; > err = ext4_buffered_write_iter(iocb, from); > if (err < 0) > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c > index fcdee27b9aa2..26b3c84d7f64 100644 > --- a/fs/ext4/inode.c > +++ b/fs/ext4/inode.c > @@ -3449,12 +3449,16 @@ static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, > { > /* > * Check to see whether an error occurred while writing out the data to > - * the allocated blocks. If so, return the magic error code so that we > - * fallback to buffered I/O and attempt to complete the remainder of > - * the I/O. Any blocks that may have been allocated in preparation for > - * the direct I/O will be reused during buffered I/O. > + * the allocated blocks. If so, return the magic error code for > + * non-atomic write so that we fallback to buffered I/O and attempt to > + * complete the remainder of the I/O. > + * For atomic writes we will simply fail the I/O request if we coudn't > + * write anything. For non-atomic writes, any blocks that may have been > + * allocated in preparation for the direct I/O will be reused during > + * buffered I/O. > */ > - if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) > + if (!(flags & IOMAP_ATOMIC) && (flags & (IOMAP_WRITE | IOMAP_DIRECT)) Huh. The WRITE|DIRECT check doesn't look right to me, because the expression returns true for any write or any directio. I think that's currently "ok" because ext4_iomap_end is only called for directio writes, but this bugs me anyway. For a directio write fallback, that comparison really should be: (flags & (WRITE|DIRECT)) == (WRITE|DIRECT) static inline bool ext4_want_directio_fallback(unsigned flags, ssize_t written) { /* must be a directio to fall back to buffered */ if (flags & (IOMAP_WRITE | IOMAP_DIRECT)) != (IOMAP_WRITE | IOMAP_DIRECT) return false; /* atomic writes are all-or-nothing */ if (flags & IOMAP_ATOMIC) return false; /* can only try again if we wrote nothing */ return written == 0; } if (ext4_want_directio_fallback(flags, written)) return -ENOTBLK; > + && written == 0) Nit: put the '&&' operator on the previous line when there's a multiline expression. --D > return -ENOTBLK; > > return 0; > -- > 2.46.0 > >
