> > > [snip]
> > > > @@ -436,8 +485,20 @@ static ssize_t exfat_direct_IO(struct kiocb
> > > > *iocb, struct iov_iter *iter)
> > > > * condition of exfat_get_block() and ->truncate().
> > > > */
> > > > ret = blockdev_direct_IO(iocb, inode, iter, exfat_get_block);
> > > > - if (ret < 0 && (rw & WRITE))
> > > > - exfat_write_failed(mapping, size);
> > > > + if (ret < 0) {
> > > > + if (rw & WRITE)
> > > > + exfat_write_failed(mapping, size);
> > > > +
> > > > + if (ret != -EIOCBQUEUED)
> > > > + return ret;
> > > > + } else
> > > > + size = pos + ret;
> > > > +
> > > > + if ((rw & READ) && pos < ei->valid_size && ei->valid_size < size) {
> > > > + iov_iter_revert(iter, size - ei->valid_size);
> > > > + iov_iter_zero(size - ei->valid_size, iter);
> > > > + }
> > >
> > > This approach causes unnecessary reads to the range after
> > > valid_size,
> > right?
> >
> > I don't think so.
> >
> > If the blocks across valid_size, the iov_iter will be handle as 1.
> > Read the blocks before valid_size.
> > 2. Read the block where valid_size is located and set the area after
> > valid_size to zero.
> > 3. zero the buffer of the blocks after valid_size(not read from disk)
> >
> > So there are unnecessary zeroing here(in 1 and 2), no unnecessary reads.
> > I will remove the unnecessary zeroing.
>
> You are right. There might be no need to change.
> It could be handled in do_direct_IO() with get_block newly modifed.
Since "size = pos + ret;" is updated to the actual read position, there is no unnecessary zeroing.
The code does not need to be updated.
