Re: [PATCH 11/15] fs: dedup file range to use a compare function

[Goldwyn Rodrigues <rgoldwyn@suse.de>]

On 10:04 28/03, Darrick J. Wong wrote:
> On Tue, Mar 26, 2019 at 02:02:57PM -0500, Goldwyn Rodrigues wrote:
> > From: Goldwyn Rodrigues <rgoldwyn@suse.com>
> > 
> > With dax we cannot deal with readpage() etc. So, we create a
> > funciton callback to perform the file data comparison and pass
> > it to generic_remap_file_range_prep() so it can use iomap-based
> > functions.
> 
> So it occurs to me -- is there an intrinsic requirement that all files
> sharing blocks must be in DAX mode or not-DAX mode?  Or in other words,
> if I run cp --reflink a b, then it cannot be the case that IS_DAX(a) !=
> IS_DAX(b), right?
> 
> Will there be calamitous consequences if dax and page cache both point
> to the same piece of storage?  I would imagine so...

Yes, data corruptions!

> 
> > This may not be the best way to solve this. Suggestions welcome.
> 
> Agree.  There's a fair amount of code duplication between the dax and
> pagecache compare functions, considering that the loop structure for
> both cases is:
> 
> while (len) {
> 	/* Figure out minimum comparison length */
> 
> 	/* Map source file range into memory */
> 
> 	/* Map dest file range into memory */
> 
> 	/* Compare memory */
> 
> 	/* Release dest file memory */
> 
> 	/* Release source file memory */
> 
> 	/* Update counters or break out */
> }
> 
> The current vfs_dedupe_file_range_compare could use some improvements,
> at least for filesystems where we actually support iomap.  In
> particular, if you try to dedupe into or out of a hole, it'll flood the
> page cache with zero pages, which is pretty wasteful if we could've
> found out that it's actually a hole.
> 
> > Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
> > ---
> >  fs/btrfs/ctree.h     |  3 +++
> >  fs/btrfs/dax.c       |  7 +++++++
> >  fs/btrfs/ioctl.c     | 13 +++++++++++-
> >  fs/dax.c             | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  fs/ocfs2/file.c      |  2 +-
> >  fs/read_write.c      |  9 +++++---
> >  fs/xfs/xfs_reflink.c |  2 +-
> >  include/linux/dax.h  |  2 ++
> >  include/linux/fs.h   |  4 +++-
> >  9 files changed, 93 insertions(+), 7 deletions(-)
> > 
> > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> > index 0e5060933bde..750f9c70fabe 100644
> > --- a/fs/btrfs/ctree.h
> > +++ b/fs/btrfs/ctree.h
> > @@ -3803,6 +3803,9 @@ int btree_readahead_hook(struct extent_buffer *eb, int err);
> >  ssize_t btrfs_file_dax_read(struct kiocb *iocb, struct iov_iter *to);
> >  ssize_t btrfs_file_dax_write(struct kiocb *iocb, struct iov_iter *from);
> >  vm_fault_t btrfs_dax_fault(struct vm_fault *vmf);
> > +int btrfs_dax_file_range_compare(struct inode *src, loff_t srcoff,
> > +		struct inode *dest, loff_t destoff, loff_t len,
> > +		bool *is_same);
> >  #else
> >  static inline ssize_t btrfs_file_dax_write(struct kiocb *iocb, struct iov_iter *from)
> >  {
> > diff --git a/fs/btrfs/dax.c b/fs/btrfs/dax.c
> > index 927f962d1e88..9488cae0f8b4 100644
> > --- a/fs/btrfs/dax.c
> > +++ b/fs/btrfs/dax.c
> > @@ -168,4 +168,11 @@ vm_fault_t btrfs_dax_fault(struct vm_fault *vmf)
> >  
> >  	return ret;
> >  }
> > +
> > +int btrfs_dax_file_range_compare(struct inode *src, loff_t srcoff,
> > +		struct inode *dest, loff_t destoff, loff_t len,
> > +		bool *is_same)
> > +{
> > +	return dax_file_range_compare(src, srcoff, dest, destoff, len, is_same, &btrfs_iomap_ops);
> > +}
> >  #endif /* CONFIG_FS_DAX */
> > diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> > index e66426e7692d..2e5137b01561 100644
> > --- a/fs/btrfs/ioctl.c
> > +++ b/fs/btrfs/ioctl.c
> > @@ -3990,8 +3990,19 @@ static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
> >  	if (ret < 0)
> >  		goto out_unlock;
> >  
> > +#ifdef CONFIG_FS_DAX
> > +	if (IS_DAX(file_inode(file_in)) && IS_DAX(file_inode(file_out)))
> > +		ret = generic_remap_file_range_prep(file_in, pos_in, file_out,
> > +				pos_out, len, remap_flags,
> > +				btrfs_dax_file_range_compare);
> > +	else
> > +		ret = generic_remap_file_range_prep(file_in, pos_in, file_out,
> > +				pos_out, len, remap_flags, NULL);
> > +#else
> >  	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
> > -					    len, remap_flags);
> > +						    len, remap_flags, NULL);
> > +#endif
> > +
> >  	if (ret < 0 || *len == 0)
> >  		goto out_unlock;
> >  
> > diff --git a/fs/dax.c b/fs/dax.c
> > index 41061da42771..18998c5ee27a 100644
> > --- a/fs/dax.c
> > +++ b/fs/dax.c
> > @@ -1775,3 +1775,61 @@ vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf,
> >  	return dax_insert_pfn_mkwrite(vmf, pfn, order);
> >  }
> >  EXPORT_SYMBOL_GPL(dax_finish_sync_fault);
> > +
> > +
> > +int dax_file_range_compare(struct inode *src, loff_t srcoff, struct inode *dest,
> > +		loff_t destoff, loff_t len, bool *is_same, const struct iomap_ops *ops)
> 
> Comments for this function assume this is a vfs generic function, not a
> dax specific thing...
> 
> > +{
> > +	void *saddr, *daddr;
> > +	struct iomap s_iomap = {0};
> > +	struct iomap d_iomap = {0};
> > +	loff_t dstart, sstart;
> > +	bool same = true;
> > +	loff_t cmp_len, l;
> > +	int id, ret = 0;
> > +
> > +	id = dax_read_lock();
> > +	while (len) {
> > +		ret = ops->iomap_begin(src, srcoff, len, 0, &s_iomap);
> > +		if (ret < 0) {
> > +			if (ops->iomap_end)
> > +				ops->iomap_end(src, srcoff, len, ret, 0, &s_iomap);
> > +			return ret;
> > +		}
> 
> Shouldn't we be checking that the iomap actually points to written
> storage?  As opposed to hole/delalloc/unwritten?
> 
> Bonus: not-written extents (hole/unwritten) could be optimized a bit,
> though I don't know if it's worth it for something that's probably a
> corner case.

Yes, I think this needs to be optimized. However, from the VFS API end.
As you mention above, it fills the pagecache unnecessarily. I feel file
comparison is best handled by individual filesystems as opposed to
the VFS.

> 
> > +		cmp_len = len;
> > +		if (cmp_len > s_iomap.offset + s_iomap.length - srcoff)
> > +			cmp_len = s_iomap.offset + s_iomap.length - srcoff;
> > +
> > +		ret = ops->iomap_begin(dest, destoff, cmp_len, 0, &d_iomap);
> > +		if (ret < 0) {
> > +			if (ops->iomap_end) {
> > +				ops->iomap_end(src, srcoff, len, ret, 0, &s_iomap);
> > +				ops->iomap_end(dest, destoff, len, ret, 0, &d_iomap);
> > +			}
> > +			return ret;
> > +		}
> > +		if (cmp_len > d_iomap.offset + d_iomap.length - destoff)
> > +			cmp_len = d_iomap.offset + d_iomap.length - destoff;
> > +
> 
> If you wanted to make this a generic function, it would be kinda nice if
> we could switch between grabbing page cache for non-DAX files and
> dax_direct_access for DAX files.
> 
> (Actually, no, that's kind of ugly too.)
> 
> Perhaps iomap.c needs to grow a function to iterate ranges of two files
> and call an actor function on both sets?  And we can change the actor to
> be the pagecache-grabbing one or the dax-access one depending on the DAX
> mode.
> 

Yes, that would work well.

> Though I guess that leaves poor old ocfs2 in the dust since it doesn't
> support iomap.  I guess it wouldn't be too hard to adapt it to have an
> iomap ops for reads only.
> 
> > +
> > +		sstart = (get_start_sect(s_iomap.bdev) << 9) + s_iomap.addr + (srcoff - s_iomap.offset);
> > +		l = dax_direct_access(s_iomap.dax_dev, PHYS_PFN(sstart), PHYS_PFN(cmp_len), &saddr, NULL);
> > +		dstart = (get_start_sect(d_iomap.bdev) << 9) + d_iomap.addr + (destoff - d_iomap.offset);
> > +		l = dax_direct_access(d_iomap.dax_dev, PHYS_PFN(dstart), PHYS_PFN(cmp_len), &daddr, NULL);
> > +		same = !memcmp(saddr, daddr, cmp_len);
> > +		if (!same)
> > +			break;
> > +		len -= cmp_len;
> > +		srcoff += cmp_len;
> > +		destoff += cmp_len;
> > +
> > +		if (ops->iomap_end) {
> > +			ret = ops->iomap_end(src, srcoff, len, 0, 0, &s_iomap);
> > +			ret = ops->iomap_end(dest, destoff, len, 0, 0, &d_iomap);
> > +		}
> > +	}
> > +	dax_read_unlock(id);
> > +	*is_same = same;
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(dax_file_range_compare);
> > diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
> > index d640c5f8a85d..6bf3e8fbb016 100644
> > --- a/fs/ocfs2/file.c
> > +++ b/fs/ocfs2/file.c
> > @@ -2558,7 +2558,7 @@ static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
> >  		goto out_unlock;
> >  
> >  	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
> > -			&len, remap_flags);
> > +			&len, remap_flags, NULL);
> >  	if (ret < 0 || len == 0)
> >  		goto out_unlock;
> >  
> > diff --git a/fs/read_write.c b/fs/read_write.c
> > index 177ccc3d405a..da521a221213 100644
> > --- a/fs/read_write.c
> > +++ b/fs/read_write.c
> > @@ -1855,7 +1855,7 @@ static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
> >   */
> >  int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
> >  				  struct file *file_out, loff_t pos_out,
> > -				  loff_t *len, unsigned int remap_flags)
> > +				  loff_t *len, unsigned int remap_flags, compare_range_t compare)
> 
> Line wrapping...
> 
> >  {
> >  	struct inode *inode_in = file_inode(file_in);
> >  	struct inode *inode_out = file_inode(file_out);
> > @@ -1914,8 +1914,11 @@ int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
> >  	 */
> >  	if (remap_flags & REMAP_FILE_DEDUP) {
> >  		bool		is_same = false;
> > -
> > -		ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
> > +		if (compare)
> > +			ret = compare(inode_in, pos_in,
> > +				inode_out, pos_out, *len, &is_same);
> > +		else
> > +			ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
> >  				inode_out, pos_out, *len, &is_same);
> 
> Make the callers pass in vfs_dedupe_file_range_compare instead of NULL,
> and avoid this if clause stuff.

Got it. Thanks.

> 
> --D
> 
> >  		if (ret)
> >  			return ret;
> > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
> > index 680ae7662a78..8907c7aa3f19 100644
> > --- a/fs/xfs/xfs_reflink.c
> > +++ b/fs/xfs/xfs_reflink.c
> > @@ -1350,7 +1350,7 @@ xfs_reflink_remap_prep(
> >  		goto out_unlock;
> >  
> >  	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
> > -			len, remap_flags);
> > +			len, remap_flags, NULL);
> >  	if (ret < 0 || *len == 0)
> >  		goto out_unlock;
> >  
> > diff --git a/include/linux/dax.h b/include/linux/dax.h
> > index 0dd316a74a29..a11bc7b1f526 100644
> > --- a/include/linux/dax.h
> > +++ b/include/linux/dax.h
> > @@ -157,6 +157,8 @@ vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf,
> >  int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index);
> >  int dax_invalidate_mapping_entry_sync(struct address_space *mapping,
> >  				      pgoff_t index);
> > +int dax_file_range_compare(struct inode *src, loff_t srcoff, struct inode *dest,
> > +                loff_t destoff, loff_t len, bool *is_same, const struct iomap_ops *ops);
> >  
> >  #ifdef CONFIG_FS_DAX
> >  int __dax_zero_page_range(struct block_device *bdev,
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index 8b42df09b04c..22fe4324b22e 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -1880,10 +1880,12 @@ extern ssize_t vfs_readv(struct file *, const struct iovec __user *,
> >  		unsigned long, loff_t *, rwf_t);
> >  extern ssize_t vfs_copy_file_range(struct file *, loff_t , struct file *,
> >  				   loff_t, size_t, unsigned int);
> > +typedef int (compare_range_t)(struct inode *src, loff_t srcpos, struct inode *dest,
> > +		loff_t destpos, loff_t len, bool *is_same);
> >  extern int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
> >  					 struct file *file_out, loff_t pos_out,
> >  					 loff_t *count,
> > -					 unsigned int remap_flags);
> > +					 unsigned int remap_flags, compare_range_t cmp);
> >  extern loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
> >  				  struct file *file_out, loff_t pos_out,
> >  				  loff_t len, unsigned int remap_flags);
> > -- 
> > 2.16.4
> > 
> 

-- 
Goldwyn