On Fri, Feb 26, 2021 at 06:16:26AM +0100, Christoph Hellwig wrote:

On Fri, Feb 26, 2021 at 10:32:50AM +1100, Stephen Rothwell wrote:

quoted

quoted

quoted

 -	return filemap_read(iocb, to, ret);
 +	if (iocb->ki_flags & IOCB_NOWAIT)
 +		iocb->ki_flags |= IOCB_NOIO;
 +
- 	ret = generic_file_buffered_read(iocb, to, ret);
++	ret = filemap_read(iocb, to, ret);
 +
 +	if (iocb->ki_flags & IOCB_NOWAIT) {
 +		iocb->ki_flags &= ~IOCB_NOIO;
 +		if (ret == 0)
 +			ret = -EAGAIN;
 +	}
 +
 +	return ret;
  }

I think the above code looks completely bogus.  Instead whatever code
in btrfs hecks for IOCB_NOIO to avoid blocking readahead should also
check IOCB_NOWAIT.

Thanks for the comment, I've removed the patch from for-next and
notified the authors.