Re: aio completions vs file_accessed race, was: Re: [PATCH 7/8] xfs: split direct I/O and DAX path

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Thu, Sep 29, 2016 at 10:18:49PM +0200, Christoph Hellwig wrote:
> On Thu, Sep 29, 2016 at 10:18:34PM +0200, Christoph Hellwig wrote:
> > Can you try the patch below?  That just moves the file_accessed call
> > before the I/O, similar to how we handle timestamp updates on the write
> > side.  generic_file_read_iter will also need a similar update.
> 
> And now with patch:
> 
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index 349f328..6919412 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -270,6 +270,8 @@ xfs_file_dio_aio_read(
>  		return -EINVAL;
>  	}
>  
> +	file_accessed(iocb->ki_filp);
> +
>  	/*
>  	 * Locking is a bit tricky here. If we take an exclusive lock for direct
>  	 * IO, we effectively serialise all new concurrent read IO to this file
> @@ -324,7 +326,6 @@ xfs_file_dio_aio_read(
>  	}
>  	xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
>  
> -	file_accessed(iocb->ki_filp);
>  	return ret;
>  }
>  

I noticed that g/323 only breaks when the XFS is on a real disk; on
pmem it works fine with or without the patch.

That said, it makes the crash go away and the patch looks ok, so:
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Tested-by: Darrick J. Wong <darrick.wong@oracle.com>

--D