Re: [PATCH 04/15] dax: Introduce IOMAP_F_COW for copy-on-write

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

On Wed, Mar 27, 2019 at 01:58:53PM -0500, Goldwyn Rodrigues wrote:
> On 10:54 27/03, Darrick J. Wong wrote:
> > On Tue, Mar 26, 2019 at 02:02:50PM -0500, Goldwyn Rodrigues wrote:
> > > From: Goldwyn Rodrigues <rgoldwyn@suse.com>
> > > 
> > > The IOMAP_F_COW is a flag to notify dax that it needs to copy
> > > the data from iomap->cow_addr to iomap->addr, if the start/end
> > > of I/O are not page aligned.
> > > 
> > > This also introduces dax_to_dax_copy() which performs a copy
> > > from one part of the device to another, to a maximum of one page.
> > > 
> > > Question: Using iomap.cow_addr == 0 means the CoW is to be copied
> > > (or memset) from a hole. Would this be better handled through a flag?
> > > 
> > > Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
> > > ---
> > >  fs/dax.c              | 36 ++++++++++++++++++++++++++++++++++++
> > >  include/linux/iomap.h |  3 +++
> > >  2 files changed, 39 insertions(+)
> > > 
> > > diff --git a/fs/dax.c b/fs/dax.c
> > > index ca0671d55aa6..e254535dd830 100644
> > > --- a/fs/dax.c
> > > +++ b/fs/dax.c
> > > @@ -1051,6 +1051,28 @@ static bool dax_range_is_aligned(struct block_device *bdev,
> > >  	return true;
> > >  }
> > >  
> > > +static void dax_to_dax_copy(struct iomap *iomap, loff_t pos, void *daddr,
> > > +			    size_t len)

Hmm... is this a dax copy-in function?  I think what's going on here is
that we have a request to copy into the file @len bytes at offset @pos;
@daddr is the pmemory address of file offset @pos, and the @iomap is
supposed to have a cow address (or nothing).

(Would love a comment here, even if it is a static helper.)

> > > +{
> > > +	loff_t blk_start, blk_pg;
> > > +	void *saddr;
> > > +	ssize_t map_len;
> > > +
> > > +	/* A zero address is a hole. */
> > > +	if (iomap->cow_addr == 0) {
> > > +		memset(daddr, 0, len);
> > > +		return;
> > > +	}
> > > +
> > > +	blk_start = iomap->cow_addr + pos - iomap->cow_pos;
> > > +	blk_pg = round_down(blk_start, PAGE_SIZE);
> > > +
> > > +	map_len = dax_direct_access(iomap->dax_dev, PHYS_PFN(blk_pg), PAGE_SIZE,
> > > +			&saddr, NULL);

/me wonders if we're supposed to do something with map_len here?

> > > +	saddr += blk_start - blk_pg;
> > > +	memcpy(daddr, saddr, len);
> > > +}
> > > +
> > >  int __dax_zero_page_range(struct block_device *bdev,
> > >  		struct dax_device *dax_dev, sector_t sector,
> > >  		unsigned int offset, unsigned int size)
> > > @@ -1143,6 +1165,20 @@ dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
> > >  			break;
> > >  		}
> > >  
> > > +		if (iomap->flags & IOMAP_F_COW) {
> > > +			loff_t pg_end = round_up(end, PAGE_SIZE);
> > > +			/*
> > > +			 * Copy the first part of the page
> > > +			 * Note: we pass offset as length
> > > +			 */
> > > +			if (offset)
> > > +				dax_to_dax_copy(iomap, pos - offset, kaddr, offset);
> > > +
> > > +			/* Copy the last part of the range */
> > > +			if (end < pg_end)
> > > +				dax_to_dax_copy(iomap, end, kaddr + offset + length, pg_end - end);
> > > +		}
> > > +
> > >  		map_len = PFN_PHYS(map_len);
> > >  		kaddr += offset;
> > >  		map_len -= offset;
> > > diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> > > index 0fefb5455bda..391785de1428 100644
> > > --- a/include/linux/iomap.h
> > > +++ b/include/linux/iomap.h
> > 
> > Probably a good idea to cc the iomap maintainers on this (entire
> > patchset)... <cough>
> 
> Yup. Will include you and Christoph in the future iterations. This will
> not be the last iteration for this patchset.
> 
> Looks like fsdevel and btrfs was too narrow a list ;)

Not necessarily, I /did/ see this on both lists.  Just a friendly
reminder that iomap has maintainers now and is no longer adrift in the
VFS. :)

> > 
> > --D
> > 
> > > @@ -35,6 +35,7 @@ struct vm_fault;
> > >  #define IOMAP_F_NEW		0x01	/* blocks have been newly allocated */
> > >  #define IOMAP_F_DIRTY		0x02	/* uncommitted metadata */
> > >  #define IOMAP_F_BUFFER_HEAD	0x04	/* file system requires buffer heads */
> > > +#define IOMAP_F_COW		0x08	/* cow before write */

This could use some more elaboration, since I couldn't figure out with
certainty how this mechanism is supposed to work...

> > >  
> > >  /*
> > >   * Flags that only need to be reported for IOMAP_REPORT requests:
> > > @@ -59,6 +60,8 @@ struct iomap {
> > >  	u64			length;	/* length of mapping, bytes */
> > >  	u16			type;	/* type of mapping */
> > >  	u16			flags;	/* flags for mapping */
> > > +	u64			cow_addr; /* read address to perform CoW */

Oh, I see, IOMAP_F_COW means @cow_addr points to the copy source
address, not the destination space we just allocated.

/*
 * A copy-on-write operation is necessary to maintain data integrity.
 * Write actors must copy the portion of the file that will not be
 * overwritten by the write from @cow_addr to @addr.
 */
#define IOMAP_F_COW		0x08

> > > +	loff_t			cow_pos; /* file offset of cow_addr */

Why wouldn't this be named @cow_offset?

And why wouldn't it be the same as @offset?

--D

> > >  	struct block_device	*bdev;	/* block device for I/O */
> > >  	struct dax_device	*dax_dev; /* dax_dev for dax operations */
> > >  	void			*inline_data;
> > > -- 
> > > 2.16.4
> > > 
> > 
> 
> -- 
> Goldwyn