Re: [PATCH v2 5/6] block: Add support for DAX reads/writes to block devices

[Boaz Harrosh <boaz@plexistor.com>]

On 07/03/2015 05:40 PM, Matthew Wilcox wrote:
> If a block device supports the ->direct_access methods, bypass the normal
> DIO path and use DAX to go straight to memcpy() instead of allocating
> a DIO and a BIO.
> 

I can't remember the details but I'm not sure it is safe for mmap to go through
page-cache while DAX is bypassing page-cache. (Because of the pg_mkwrite thing,
while IO), do you find this code safe?

I think you need to force all DAX IO through dax.c. I did not understand why we
need to give the user an option? why would he ever want a cached access to a memory
device (ie two copies of the same thing). Why not DAX hard coded?

> Includes support for the DIO_SKIP_DIO_COUNT flag in DAX, as is done in
> do_blockdev_direct_IO().
> 
> Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
> ---
>  fs/block_dev.c | 4 ++++
>  fs/dax.c       | 6 ++++--
>  2 files changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 4fe10f9..0bb2993 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -152,6 +152,9 @@ blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
>  	struct file *file = iocb->ki_filp;
>  	struct inode *inode = file->f_mapping->host;
>  
> +	if (IS_DAX(inode))
> +		return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
> +				NULL, DIO_SKIP_DIO_COUNT);
>  	return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
>  				    blkdev_get_block, NULL, NULL,
>  				    DIO_SKIP_DIO_COUNT);
> @@ -1170,6 +1173,7 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
>  		bdev->bd_disk = disk;
>  		bdev->bd_queue = disk->queue;
>  		bdev->bd_contains = bdev;
> +		bdev->bd_inode->i_flags = disk->fops->direct_access ? S_DAX : 0;
>  		if (!partno) {
>  			ret = -ENXIO;
>  			bdev->bd_part = disk_get_part(disk, partno);
> diff --git a/fs/dax.c b/fs/dax.c
> index eaa9e06..c3e21cc 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -209,7 +209,8 @@ ssize_t dax_do_io(struct kiocb *iocb, struct inode *inode,
>  	}
>  
>  	/* Protects against truncate */
> -	inode_dio_begin(inode);
> +	if (!(flags & DIO_SKIP_DIO_COUNT))
> +		inode_dio_begin(inode);

Is really a separate issue, is it not?

>  
>  	retval = dax_io(inode, iter, pos, end, get_block, &bh);
>  
> @@ -219,7 +220,8 @@ ssize_t dax_do_io(struct kiocb *iocb, struct inode *inode,
>  	if ((retval > 0) && end_io)
>  		end_io(iocb, pos, retval, bh.b_private);
>  
> -	inode_dio_end(inode);
> +	if (!(flags & DIO_SKIP_DIO_COUNT))
> +		inode_dio_end(inode);
>   out:
>  	return retval;
>  }
> 

This scares me, the mix of DAX while cached data. I'd vote for hard-coded DAX
all over.

Boaz