1 files changed, 862 insertions, 608 deletions

diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
index c5b4fa004ca4..3f177b4ec131 100644
--- a/fs/xfs/xfs_reflink.c
+++ b/fs/xfs/xfs_reflink.c
@@ -11,21 +11,12 @@
 #include "xfs_trans_resv.h"
 #include "xfs_mount.h"
 #include "xfs_defer.h"
-#include "xfs_da_format.h"
-#include "xfs_da_btree.h"
 #include "xfs_inode.h"
 #include "xfs_trans.h"
-#include "xfs_inode_item.h"
 #include "xfs_bmap.h"
 #include "xfs_bmap_util.h"
-#include "xfs_error.h"
-#include "xfs_dir2.h"
-#include "xfs_dir2_priv.h"
-#include "xfs_ioctl.h"
 #include "xfs_trace.h"
-#include "xfs_log.h"
 #include "xfs_icache.h"
-#include "xfs_pnfs.h"
 #include "xfs_btree.h"
 #include "xfs_refcount_btree.h"
 #include "xfs_refcount.h"
@@ -33,13 +24,16 @@
 #include "xfs_trans_space.h"
 #include "xfs_bit.h"
 #include "xfs_alloc.h"
-#include "xfs_quota_defs.h"
 #include "xfs_quota.h"
 #include "xfs_reflink.h"
 #include "xfs_iomap.h"
-#include "xfs_rmap_btree.h"
-#include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_ag_resv.h"
+#include "xfs_health.h"
+#include "xfs_rtrefcount_btree.h"
+#include "xfs_rtalloc.h"
+#include "xfs_rtgroup.h"
+#include "xfs_metafile.h"
 
 /*
  * Copy on Write of Shared Blocks
@@ -130,41 +124,93 @@
  */
 
 /*
- * Given an AG extent, find the lowest-numbered run of shared blocks
- * within that range and return the range in fbno/flen.  If
- * find_end_of_shared is true, return the longest contiguous extent of
- * shared blocks.  If there are no shared extents, fbno and flen will
- * be set to NULLAGBLOCK and 0, respectively.
+ * Given a file mapping for the data device, find the lowest-numbered run of
+ * shared blocks within that mapping and return it in shared_offset/shared_len.
+ * The offset is relative to the start of irec.
+ *
+ * If find_end_of_shared is true, return the longest contiguous extent of shared
+ * blocks.  If there are no shared extents, shared_offset and shared_len will be
+ * set to 0;
  */
-int
+static int
 xfs_reflink_find_shared(
 	struct xfs_mount	*mp,
 	struct xfs_trans	*tp,
-	xfs_agnumber_t		agno,
-	xfs_agblock_t		agbno,
-	xfs_extlen_t		aglen,
-	xfs_agblock_t		*fbno,
-	xfs_extlen_t		*flen,
+	const struct xfs_bmbt_irec *irec,
+	xfs_extlen_t		*shared_offset,
+	xfs_extlen_t		*shared_len,
 	bool			find_end_of_shared)
 {
 	struct xfs_buf		*agbp;
+	struct xfs_perag	*pag;
 	struct xfs_btree_cur	*cur;
 	int			error;
+	xfs_agblock_t		orig_bno, found_bno;
+
+	pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, irec->br_startblock));
+	orig_bno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
 
-	error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
+	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
 	if (error)
-		return error;
-	if (!agbp)
-		return -ENOMEM;
+		goto out;
+
+	cur = xfs_refcountbt_init_cursor(mp, tp, agbp, pag);
+	error = xfs_refcount_find_shared(cur, orig_bno, irec->br_blockcount,
+			&found_bno, shared_len, find_end_of_shared);
+	xfs_btree_del_cursor(cur, error);
+	xfs_trans_brelse(tp, agbp);
+
+	if (!error && *shared_len)
+		*shared_offset = found_bno - orig_bno;
+out:
+	xfs_perag_put(pag);
+	return error;
+}
+
+/*
+ * Given a file mapping for the rt device, find the lowest-numbered run of
+ * shared blocks within that mapping and return it in shared_offset/shared_len.
+ * The offset is relative to the start of irec.
+ *
+ * If find_end_of_shared is true, return the longest contiguous extent of shared
+ * blocks.  If there are no shared extents, shared_offset and shared_len will be
+ * set to 0;
+ */
+static int
+xfs_reflink_find_rtshared(
+	struct xfs_mount	*mp,
+	struct xfs_trans	*tp,
+	const struct xfs_bmbt_irec *irec,
+	xfs_extlen_t		*shared_offset,
+	xfs_extlen_t		*shared_len,
+	bool			find_end_of_shared)
+{
+	struct xfs_rtgroup	*rtg;
+	struct xfs_btree_cur	*cur;
+	xfs_rgblock_t		orig_bno;
+	xfs_agblock_t		found_bno;
+	int			error;
 
-	cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
+	BUILD_BUG_ON(NULLRGBLOCK != NULLAGBLOCK);
 
-	error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
-			find_end_of_shared);
+	/*
+	 * Note: this uses the not quite correct xfs_agblock_t type because
+	 * xfs_refcount_find_shared is shared between the RT and data device
+	 * refcount code.
+	 */
+	orig_bno = xfs_rtb_to_rgbno(mp, irec->br_startblock);
+	rtg = xfs_rtgroup_get(mp, xfs_rtb_to_rgno(mp, irec->br_startblock));
 
+	xfs_rtgroup_lock(rtg, XFS_RTGLOCK_REFCOUNT);
+	cur = xfs_rtrefcountbt_init_cursor(tp, rtg);
+	error = xfs_refcount_find_shared(cur, orig_bno, irec->br_blockcount,
+			&found_bno, shared_len, find_end_of_shared);
 	xfs_btree_del_cursor(cur, error);
+	xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_REFCOUNT);
+	xfs_rtgroup_put(rtg);
 
-	xfs_trans_brelse(tp, agbp);
+	if (!error && *shared_len)
+		*shared_offset = found_bno - orig_bno;
 	return error;
 }
 
@@ -184,143 +230,105 @@ xfs_reflink_trim_around_shared(
 	struct xfs_bmbt_irec	*irec,
 	bool			*shared)
 {
-	xfs_agnumber_t		agno;
-	xfs_agblock_t		agbno;
-	xfs_extlen_t		aglen;
-	xfs_agblock_t		fbno;
-	xfs_extlen_t		flen;
+	struct xfs_mount	*mp = ip->i_mount;
+	xfs_extlen_t		shared_offset, shared_len;
 	int			error = 0;
 
 	/* Holes, unwritten, and delalloc extents cannot be shared */
-	if (!xfs_is_reflink_inode(ip) || !xfs_bmap_is_real_extent(irec)) {
+	if (!xfs_is_reflink_inode(ip) || !xfs_bmap_is_written_extent(irec)) {
 		*shared = false;
 		return 0;
 	}
 
 	trace_xfs_reflink_trim_around_shared(ip, irec);
 
-	agno = XFS_FSB_TO_AGNO(ip->i_mount, irec->br_startblock);
-	agbno = XFS_FSB_TO_AGBNO(ip->i_mount, irec->br_startblock);
-	aglen = irec->br_blockcount;
-
-	error = xfs_reflink_find_shared(ip->i_mount, NULL, agno, agbno,
-			aglen, &fbno, &flen, true);
+	if (XFS_IS_REALTIME_INODE(ip))
+		error = xfs_reflink_find_rtshared(mp, NULL, irec,
+				&shared_offset, &shared_len, true);
+	else
+		error = xfs_reflink_find_shared(mp, NULL, irec,
+				&shared_offset, &shared_len, true);
 	if (error)
 		return error;
 
-	*shared = false;
-	if (fbno == NULLAGBLOCK) {
+	if (!shared_len) {
 		/* No shared blocks at all. */
-		return 0;
-	} else if (fbno == agbno) {
+		*shared = false;
+	} else if (!shared_offset) {
 		/*
-		 * The start of this extent is shared.  Truncate the
-		 * mapping at the end of the shared region so that a
-		 * subsequent iteration starts at the start of the
-		 * unshared region.
+		 * The start of this mapping points to shared space.  Truncate
+		 * the mapping at the end of the shared region so that a
+		 * subsequent iteration starts at the start of the unshared
+		 * region.
 		 */
-		irec->br_blockcount = flen;
+		irec->br_blockcount = shared_len;
 		*shared = true;
-		return 0;
 	} else {
 		/*
-		 * There's a shared extent midway through this extent.
-		 * Truncate the mapping at the start of the shared
-		 * extent so that a subsequent iteration starts at the
+		 * There's a shared region that doesn't start at the beginning
+		 * of the mapping.  Truncate the mapping at the start of the
+		 * shared extent so that a subsequent iteration starts at the
 		 * start of the shared region.
 		 */
-		irec->br_blockcount = fbno - agbno;
-		return 0;
+		irec->br_blockcount = shared_offset;
+		*shared = false;
 	}
+	return 0;
 }
 
-/*
- * Trim the passed in imap to the next shared/unshared extent boundary, and
- * if imap->br_startoff points to a shared extent reserve space for it in the
- * COW fork.
- *
- * Note that imap will always contain the block numbers for the existing blocks
- * in the data fork, as the upper layers need them for read-modify-write
- * operations.
- */
 int
-xfs_reflink_reserve_cow(
+xfs_bmap_trim_cow(
 	struct xfs_inode	*ip,
-	struct xfs_bmbt_irec	*imap)
+	struct xfs_bmbt_irec	*imap,
+	bool			*shared)
 {
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
-	struct xfs_bmbt_irec	got;
-	int			error = 0;
-	bool			eof = false;
-	struct xfs_iext_cursor	icur;
-	bool			shared;
-
-	/*
-	 * Search the COW fork extent list first.  This serves two purposes:
-	 * first this implement the speculative preallocation using cowextisze,
-	 * so that we also unshared block adjacent to shared blocks instead
-	 * of just the shared blocks themselves.  Second the lookup in the
-	 * extent list is generally faster than going out to the shared extent
-	 * tree.
-	 */
-
-	if (!xfs_iext_lookup_extent(ip, ifp, imap->br_startoff, &icur, &got))
-		eof = true;
-	if (!eof && got.br_startoff <= imap->br_startoff) {
-		trace_xfs_reflink_cow_found(ip, imap);
-		xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
+	/* We can't update any real extents in always COW mode. */
+	if (xfs_is_always_cow_inode(ip) &&
+	    !isnullstartblock(imap->br_startblock)) {
+		*shared = true;
 		return 0;
 	}
 
 	/* Trim the mapping to the nearest shared extent boundary. */
-	error = xfs_reflink_trim_around_shared(ip, imap, &shared);
-	if (error)
-		return error;
-
-	/* Not shared?  Just report the (potentially capped) extent. */
-	if (!shared)
-		return 0;
-
-	/*
-	 * Fork all the shared blocks from our write offset until the end of
-	 * the extent.
-	 */
-	error = xfs_qm_dqattach_locked(ip, false);
-	if (error)
-		return error;
-
-	error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, imap->br_startoff,
-			imap->br_blockcount, 0, &got, &icur, eof);
-	if (error == -ENOSPC || error == -EDQUOT)
-		trace_xfs_reflink_cow_enospc(ip, imap);
-	if (error)
-		return error;
-
-	xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
-	trace_xfs_reflink_cow_alloc(ip, &got);
-	return 0;
+	return xfs_reflink_trim_around_shared(ip, imap, shared);
 }
 
-/* Convert part of an unwritten CoW extent to a real one. */
-STATIC int
-xfs_reflink_convert_cow_extent(
-	struct xfs_inode		*ip,
-	struct xfs_bmbt_irec		*imap,
-	xfs_fileoff_t			offset_fsb,
-	xfs_filblks_t			count_fsb)
+int
+xfs_reflink_convert_cow_locked(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		offset_fsb,
+	xfs_filblks_t		count_fsb)
 {
-	int				nimaps = 1;
+	struct xfs_iext_cursor	icur;
+	struct xfs_bmbt_irec	got;
+	struct xfs_btree_cur	*dummy_cur = NULL;
+	int			dummy_logflags;
+	int			error = 0;
 
-	if (imap->br_state == XFS_EXT_NORM)
+	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got))
 		return 0;
 
-	xfs_trim_extent(imap, offset_fsb, count_fsb);
-	trace_xfs_reflink_convert_cow(ip, imap);
-	if (imap->br_blockcount == 0)
-		return 0;
-	return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount,
-			XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, 0, imap,
-			&nimaps);
+	do {
+		if (got.br_startoff >= offset_fsb + count_fsb)
+			break;
+		if (got.br_state == XFS_EXT_NORM)
+			continue;
+		if (WARN_ON_ONCE(isnullstartblock(got.br_startblock)))
+			return -EIO;
+
+		xfs_trim_extent(&got, offset_fsb, count_fsb);
+		if (!got.br_blockcount)
+			continue;
+
+		got.br_state = XFS_EXT_NORM;
+		error = xfs_bmap_add_extent_unwritten_real(NULL, ip,
+				XFS_COW_FORK, &icur, &dummy_cur, &got,
+				&dummy_logflags);
+		if (error)
+			return error;
+	} while (xfs_iext_next_extent(ip->i_cowfp, &icur, &got));
+
+	return error;
 }
 
 /* Convert all of the unwritten CoW extents in a file's range to real ones. */
@@ -334,15 +342,12 @@ xfs_reflink_convert_cow(
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
 	xfs_filblks_t		count_fsb = end_fsb - offset_fsb;
-	struct xfs_bmbt_irec	imap;
-	int			nimaps = 1, error = 0;
+	int			error;
 
 	ASSERT(count != 0);
 
 	xfs_ilock(ip, XFS_ILOCK_EXCL);
-	error = xfs_bmapi_write(NULL, ip, offset_fsb, count_fsb,
-			XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT |
-			XFS_BMAPI_CONVERT_ONLY, 0, &imap, &nimaps);
+	error = xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;
 }
@@ -356,13 +361,13 @@ static int
 xfs_find_trim_cow_extent(
 	struct xfs_inode	*ip,
 	struct xfs_bmbt_irec	*imap,
+	struct xfs_bmbt_irec	*cmap,
 	bool			*shared,
 	bool			*found)
 {
 	xfs_fileoff_t		offset_fsb = imap->br_startoff;
 	xfs_filblks_t		count_fsb = imap->br_blockcount;
 	struct xfs_iext_cursor	icur;
-	struct xfs_bmbt_irec	got;
 
 	*found = false;
 
@@ -370,117 +375,239 @@ xfs_find_trim_cow_extent(
 	 * If we don't find an overlapping extent, trim the range we need to
 	 * allocate to fit the hole we found.
 	 */
-	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got))
-		got.br_startoff = offset_fsb + count_fsb;
-	if (got.br_startoff > offset_fsb) {
+	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, cmap))
+		cmap->br_startoff = offset_fsb + count_fsb;
+	if (cmap->br_startoff > offset_fsb) {
 		xfs_trim_extent(imap, imap->br_startoff,
-				got.br_startoff - imap->br_startoff);
-		return xfs_reflink_trim_around_shared(ip, imap, shared);
+				cmap->br_startoff - imap->br_startoff);
+		return xfs_bmap_trim_cow(ip, imap, shared);
 	}
 
 	*shared = true;
-	if (isnullstartblock(got.br_startblock)) {
-		xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
+	if (isnullstartblock(cmap->br_startblock)) {
+		xfs_trim_extent(imap, cmap->br_startoff, cmap->br_blockcount);
 		return 0;
 	}
 
 	/* real extent found - no need to allocate */
-	xfs_trim_extent(&got, offset_fsb, count_fsb);
-	*imap = got;
+	xfs_trim_extent(cmap, offset_fsb, count_fsb);
 	*found = true;
 	return 0;
 }
 
-/* Allocate all CoW reservations covering a range of blocks in a file. */
-int
-xfs_reflink_allocate_cow(
+static int
+xfs_reflink_convert_unwritten(
 	struct xfs_inode	*ip,
 	struct xfs_bmbt_irec	*imap,
-	bool			*shared,
-	uint			*lockmode)
+	struct xfs_bmbt_irec	*cmap,
+	bool			convert_now)
 {
-	struct xfs_mount	*mp = ip->i_mount;
 	xfs_fileoff_t		offset_fsb = imap->br_startoff;
 	xfs_filblks_t		count_fsb = imap->br_blockcount;
-	struct xfs_trans	*tp;
-	int			nimaps, error = 0;
-	bool			found;
-	xfs_filblks_t		resaligned;
-	xfs_extlen_t		resblks = 0;
+	int			error;
 
-	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-	ASSERT(xfs_is_reflink_inode(ip));
+	/*
+	 * cmap might larger than imap due to cowextsize hint.
+	 */
+	xfs_trim_extent(cmap, offset_fsb, count_fsb);
 
-	error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
-	if (error || !*shared)
-		return error;
-	if (found)
-		goto convert;
+	/*
+	 * COW fork extents are supposed to remain unwritten until we're ready
+	 * to initiate a disk write.  For direct I/O we are going to write the
+	 * data and need the conversion, but for buffered writes we're done.
+	 */
+	if (!convert_now || cmap->br_state == XFS_EXT_NORM)
+		return 0;
+
+	trace_xfs_reflink_convert_cow(ip, cmap);
+
+	error = xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);
+	if (!error)
+		cmap->br_state = XFS_EXT_NORM;
+
+	return error;
+}
+
+static int
+xfs_reflink_fill_cow_hole(
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*imap,
+	struct xfs_bmbt_irec	*cmap,
+	bool			*shared,
+	uint			*lockmode,
+	bool			convert_now)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	xfs_filblks_t		resaligned;
+	unsigned int		dblocks = 0, rblocks = 0;
+	int			nimaps;
+	int			error;
+	bool			found;
 
 	resaligned = xfs_aligned_fsb_count(imap->br_startoff,
 		imap->br_blockcount, xfs_get_cowextsz_hint(ip));
-	resblks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
+	if (XFS_IS_REALTIME_INODE(ip)) {
+		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
+		rblocks = resaligned;
+	} else {
+		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
+		rblocks = 0;
+	}
 
 	xfs_iunlock(ip, *lockmode);
-	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
-	*lockmode = XFS_ILOCK_EXCL;
-	xfs_ilock(ip, *lockmode);
+	*lockmode = 0;
 
+	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
+			rblocks, false, &tp);
 	if (error)
 		return error;
 
-	error = xfs_qm_dqattach_locked(ip, false);
-	if (error)
-		goto out_trans_cancel;
+	*lockmode = XFS_ILOCK_EXCL;
 
-	/*
-	 * Check for an overlapping extent again now that we dropped the ilock.
-	 */
-	error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
+	error = xfs_find_trim_cow_extent(ip, imap, cmap, shared, &found);
 	if (error || !*shared)
 		goto out_trans_cancel;
+
 	if (found) {
 		xfs_trans_cancel(tp);
 		goto convert;
 	}
 
-	error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0,
-			XFS_QMOPT_RES_REGBLKS);
-	if (error)
-		goto out_trans_cancel;
-
-	xfs_trans_ijoin(tp, ip, 0);
-
 	/* Allocate the entire reservation as unwritten blocks. */
 	nimaps = 1;
 	error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount,
-			XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC,
-			resblks, imap, &nimaps);
+			XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, 0, cmap,
+			&nimaps);
 	if (error)
-		goto out_unreserve;
+		goto out_trans_cancel;
 
 	xfs_inode_set_cowblocks_tag(ip);
 	error = xfs_trans_commit(tp);
 	if (error)
 		return error;
 
-	/*
-	 * Allocation succeeded but the requested range was not even partially
-	 * satisfied?  Bail out!
-	 */
-	if (nimaps == 0)
-		return -ENOSPC;
 convert:
-	return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb);
+	return xfs_reflink_convert_unwritten(ip, imap, cmap, convert_now);
 
-out_unreserve:
-	xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0,
-			XFS_QMOPT_RES_REGBLKS);
 out_trans_cancel:
 	xfs_trans_cancel(tp);
 	return error;
 }
 
+static int
+xfs_reflink_fill_delalloc(
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*imap,
+	struct xfs_bmbt_irec	*cmap,
+	bool			*shared,
+	uint			*lockmode,
+	bool			convert_now)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	int			nimaps;
+	int			error;
+	bool			found;
+
+	do {
+		xfs_iunlock(ip, *lockmode);
+		*lockmode = 0;
+
+		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, 0, 0,
+				false, &tp);
+		if (error)
+			return error;
+
+		*lockmode = XFS_ILOCK_EXCL;
+
+		error = xfs_find_trim_cow_extent(ip, imap, cmap, shared,
+				&found);
+		if (error || !*shared)
+			goto out_trans_cancel;
+
+		if (found) {
+			xfs_trans_cancel(tp);
+			break;
+		}
+
+		ASSERT(isnullstartblock(cmap->br_startblock) ||
+		       cmap->br_startblock == DELAYSTARTBLOCK);
+
+		/*
+		 * Replace delalloc reservation with an unwritten extent.
+		 */
+		nimaps = 1;
+		error = xfs_bmapi_write(tp, ip, cmap->br_startoff,
+				cmap->br_blockcount,
+				XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, 0,
+				cmap, &nimaps);
+		if (error)
+			goto out_trans_cancel;
+
+		xfs_inode_set_cowblocks_tag(ip);
+		error = xfs_trans_commit(tp);
+		if (error)
+			return error;
+	} while (cmap->br_startoff + cmap->br_blockcount <= imap->br_startoff);
+
+	return xfs_reflink_convert_unwritten(ip, imap, cmap, convert_now);
+
+out_trans_cancel:
+	xfs_trans_cancel(tp);
+	return error;
+}
+
+/* Allocate all CoW reservations covering a range of blocks in a file. */
+int
+xfs_reflink_allocate_cow(
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*imap,
+	struct xfs_bmbt_irec	*cmap,
+	bool			*shared,
+	uint			*lockmode,
+	bool			convert_now)
+{
+	int			error;
+	bool			found;
+
+	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
+	if (!ip->i_cowfp) {
+		ASSERT(!xfs_is_reflink_inode(ip));
+		xfs_ifork_init_cow(ip);
+	}
+
+	error = xfs_find_trim_cow_extent(ip, imap, cmap, shared, &found);
+	if (error || !*shared)
+		return error;
+
+	/* CoW fork has a real extent */
+	if (found)
+		return xfs_reflink_convert_unwritten(ip, imap, cmap,
+				convert_now);
+
+	/*
+	 * CoW fork does not have an extent and data extent is shared.
+	 * Allocate a real extent in the CoW fork.
+	 */
+	if (cmap->br_startoff > imap->br_startoff)
+		return xfs_reflink_fill_cow_hole(ip, imap, cmap, shared,
+				lockmode, convert_now);
+
+	/*
+	 * CoW fork has a delalloc reservation. Replace it with a real extent.
+	 * There may or may not be a data fork mapping.
+	 */
+	if (isnullstartblock(cmap->br_startblock) ||
+	    cmap->br_startblock == DELAYSTARTBLOCK)
+		return xfs_reflink_fill_delalloc(ip, imap, cmap, shared,
+				lockmode, convert_now);
+
+	/* Shouldn't get here. */
+	ASSERT(0);
+	return -EFSCORRUPTED;
+}
+
 /*
  * Cancel CoW reservations for some block range of an inode.
  *
@@ -498,9 +625,10 @@ xfs_reflink_cancel_cow_blocks(
 	xfs_fileoff_t			end_fsb,
 	bool				cancel_real)
 {
-	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
+	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
 	struct xfs_bmbt_irec		got, del;
 	struct xfs_iext_cursor		icur;
+	bool				isrt = XFS_IS_REALTIME_INODE(ip);
 	int				error = 0;
 
 	if (!xfs_inode_has_cow_data(ip))
@@ -522,22 +650,22 @@ xfs_reflink_cancel_cow_blocks(
 		trace_xfs_reflink_cancel_cow(ip, &del);
 
 		if (isnullstartblock(del.br_startblock)) {
-			error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK,
-					&icur, &got, &del);
-			if (error)
-				break;
+			xfs_bmap_del_extent_delay(ip, XFS_COW_FORK, &icur, &got,
+					&del, 0);
 		} else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) {
-			ASSERT((*tpp)->t_firstblock == NULLFSBLOCK);
+			ASSERT((*tpp)->t_highest_agno == NULLAGNUMBER);
 
 			/* Free the CoW orphan record. */
-			error = xfs_refcount_free_cow_extent(*tpp,
+			xfs_refcount_free_cow_extent(*tpp, isrt,
 					del.br_startblock, del.br_blockcount);
+
+			error = xfs_free_extent_later(*tpp, del.br_startblock,
+					del.br_blockcount, NULL,
+					XFS_AG_RESV_NONE,
+					isrt ? XFS_FREE_EXTENT_REALTIME : 0);
 			if (error)
 				break;
 
-			xfs_bmap_add_free(*tpp, del.br_startblock,
-					  del.br_blockcount, NULL);
-
 			/* Roll the transaction */
 			error = xfs_defer_finish(tpp);
 			if (error)
@@ -547,11 +675,7 @@ xfs_reflink_cancel_cow_blocks(
 			xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
 
 			/* Remove the quota reservation */
-			error = xfs_trans_reserve_quota_nblks(NULL, ip,
-					-(long)del.br_blockcount, 0,
-					XFS_QMOPT_RES_REGBLKS);
-			if (error)
-				break;
+			xfs_quota_unreserve_blkres(ip, del.br_blockcount);
 		} else {
 			/* Didn't do anything, push cursor back. */
 			xfs_iext_prev(ifp, &icur);
@@ -586,7 +710,7 @@ xfs_reflink_cancel_cow_range(
 	int			error;
 
 	trace_xfs_reflink_cancel_cow_range(ip, offset, count);
-	ASSERT(xfs_is_reflink_inode(ip));
+	ASSERT(ip->i_cowfp);
 
 	offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
 	if (count == NULLFILEOFF)
@@ -596,7 +720,7 @@ xfs_reflink_cancel_cow_range(
 
 	/* Start a rolling transaction to remove the mappings */
 	error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
-			0, 0, XFS_TRANS_NOFS, &tp);
+			0, 0, 0, &tp);
 	if (error)
 		goto out;
 
@@ -622,6 +746,35 @@ out:
 	return error;
 }
 
+#ifdef CONFIG_XFS_QUOTA
+/*
+ * Update quota accounting for a remapping operation.  When we're remapping
+ * something from the CoW fork to the data fork, we must update the quota
+ * accounting for delayed allocations.  For remapping from the data fork to the
+ * data fork, use regular block accounting.
+ */
+static inline void
+xfs_reflink_update_quota(
+	struct xfs_trans	*tp,
+	struct xfs_inode	*ip,
+	bool			is_cow,
+	int64_t			blocks)
+{
+	unsigned int		qflag;
+
+	if (XFS_IS_REALTIME_INODE(ip)) {
+		qflag = is_cow ? XFS_TRANS_DQ_DELRTBCOUNT :
+				 XFS_TRANS_DQ_RTBCOUNT;
+	} else {
+		qflag = is_cow ? XFS_TRANS_DQ_DELBCOUNT :
+				 XFS_TRANS_DQ_BCOUNT;
+	}
+	xfs_trans_mod_dquot_byino(tp, ip, qflag, blocks);
+}
+#else
+# define xfs_reflink_update_quota(tp, ip, is_cow, blocks)	((void)0)
+#endif
+
 /*
  * Remap part of the CoW fork into the data fork.
  *
@@ -633,111 +786,142 @@ out:
  * requirements as low as possible.
  */
 STATIC int
-xfs_reflink_end_cow_extent(
+xfs_reflink_end_cow_extent_locked(
+	struct xfs_trans	*tp,
 	struct xfs_inode	*ip,
-	xfs_fileoff_t		offset_fsb,
-	xfs_fileoff_t		*end_fsb)
+	xfs_fileoff_t		*offset_fsb,
+	xfs_fileoff_t		end_fsb)
 {
-	struct xfs_bmbt_irec	got, del;
 	struct xfs_iext_cursor	icur;
-	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_trans	*tp;
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
-	xfs_filblks_t		rlen;
-	unsigned int		resblks;
+	struct xfs_bmbt_irec	got, del, data;
+	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
+	int			nmaps;
+	bool			isrt = XFS_IS_REALTIME_INODE(ip);
 	int			error;
 
-	/* No COW extents?  That's easy! */
-	if (ifp->if_bytes == 0) {
-		*end_fsb = offset_fsb;
-		return 0;
-	}
-
-	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
-	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
-			XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
-	if (error)
-		return error;
-
-	/*
-	 * Lock the inode.  We have to ijoin without automatic unlock because
-	 * the lead transaction is the refcountbt record deletion; the data
-	 * fork update follows as a deferred log item.
-	 */
-	xfs_ilock(ip, XFS_ILOCK_EXCL);
-	xfs_trans_ijoin(tp, ip, 0);
-
 	/*
 	 * In case of racing, overlapping AIO writes no COW extents might be
 	 * left by the time I/O completes for the loser of the race.  In that
 	 * case we are done.
 	 */
-	if (!xfs_iext_lookup_extent_before(ip, ifp, end_fsb, &icur, &got) ||
-	    got.br_startoff + got.br_blockcount <= offset_fsb) {
-		*end_fsb = offset_fsb;
-		goto out_cancel;
+	if (!xfs_iext_lookup_extent(ip, ifp, *offset_fsb, &icur, &got) ||
+	    got.br_startoff >= end_fsb) {
+		*offset_fsb = end_fsb;
+		return 0;
 	}
 
 	/*
-	 * Structure copy @got into @del, then trim @del to the range that we
-	 * were asked to remap.  We preserve @got for the eventual CoW fork
+	 * Only remap real extents that contain data.  With AIO, speculative
+	 * preallocations can leak into the range we are called upon, and we
+	 * need to skip them.  Preserve @got for the eventual CoW fork
 	 * deletion; from now on @del represents the mapping that we're
 	 * actually remapping.
 	 */
+	while (!xfs_bmap_is_written_extent(&got)) {
+		if (!xfs_iext_next_extent(ifp, &icur, &got) ||
+		    got.br_startoff >= end_fsb) {
+			*offset_fsb = end_fsb;
+			return 0;
+		}
+	}
 	del = got;
-	xfs_trim_extent(&del, offset_fsb, *end_fsb - offset_fsb);
-
-	ASSERT(del.br_blockcount > 0);
+	xfs_trim_extent(&del, *offset_fsb, end_fsb - *offset_fsb);
 
-	/*
-	 * Only remap real extents that contain data.  With AIO, speculative
-	 * preallocations can leak into the range we are called upon, and we
-	 * need to skip them.
-	 */
-	if (!xfs_bmap_is_real_extent(&got)) {
-		*end_fsb = del.br_startoff;
-		goto out_cancel;
-	}
+	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
+			XFS_IEXT_REFLINK_END_COW_CNT);
+	if (error)
+		return error;
 
-	/* Unmap the old blocks in the data fork. */
-	rlen = del.br_blockcount;
-	error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1);
+	/* Grab the corresponding mapping in the data fork. */
+	nmaps = 1;
+	error = xfs_bmapi_read(ip, del.br_startoff, del.br_blockcount, &data,
+			&nmaps, 0);
 	if (error)
-		goto out_cancel;
+		return error;
+
+	/* We can only remap the smaller of the two extent sizes. */
+	data.br_blockcount = min(data.br_blockcount, del.br_blockcount);
+	del.br_blockcount = data.br_blockcount;
+
+	trace_xfs_reflink_cow_remap_from(ip, &del);
+	trace_xfs_reflink_cow_remap_to(ip, &data);
 
-	/* Trim the extent to whatever got unmapped. */
-	xfs_trim_extent(&del, del.br_startoff + rlen, del.br_blockcount - rlen);
-	trace_xfs_reflink_cow_remap(ip, &del);
+	if (xfs_bmap_is_real_extent(&data)) {
+		/*
+		 * If the extent we're remapping is backed by storage (written
+		 * or not), unmap the extent and drop its refcount.
+		 */
+		xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &data);
+		xfs_refcount_decrease_extent(tp, isrt, &data);
+		xfs_reflink_update_quota(tp, ip, false, -data.br_blockcount);
+	} else if (data.br_startblock == DELAYSTARTBLOCK) {
+		int		done;
+
+		/*
+		 * If the extent we're remapping is a delalloc reservation,
+		 * we can use the regular bunmapi function to release the
+		 * incore state.  Dropping the delalloc reservation takes care
+		 * of the quota reservation for us.
+		 */
+		error = xfs_bunmapi(NULL, ip, data.br_startoff,
+				data.br_blockcount, 0, 1, &done);
+		if (error)
+			return error;
+		ASSERT(done);
+	}
 
 	/* Free the CoW orphan record. */
-	error = xfs_refcount_free_cow_extent(tp, del.br_startblock,
+	xfs_refcount_free_cow_extent(tp, isrt, del.br_startblock,
 			del.br_blockcount);
-	if (error)
-		goto out_cancel;
 
 	/* Map the new blocks into the data fork. */
-	error = xfs_bmap_map_extent(tp, ip, &del);
-	if (error)
-		goto out_cancel;
+	xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, &del);
 
 	/* Charge this new data fork mapping to the on-disk quota. */
-	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_DELBCOUNT,
-			(long)del.br_blockcount);
+	xfs_reflink_update_quota(tp, ip, true, del.br_blockcount);
 
 	/* Remove the mapping from the CoW fork. */
 	xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
 
-	error = xfs_trans_commit(tp);
-	xfs_iunlock(ip, XFS_ILOCK_EXCL);
-	if (error)
-		return error;
-
 	/* Update the caller about how much progress we made. */
-	*end_fsb = del.br_startoff;
+	*offset_fsb = del.br_startoff + del.br_blockcount;
 	return 0;
+}
 
-out_cancel:
-	xfs_trans_cancel(tp);
+/*
+ * Remap part of the CoW fork into the data fork.
+ *
+ * We aim to remap the range starting at @offset_fsb and ending at @end_fsb
+ * into the data fork; this function will remap what it can (at the end of the
+ * range) and update @end_fsb appropriately.  Each remap gets its own
+ * transaction because we can end up merging and splitting bmbt blocks for
+ * every remap operation and we'd like to keep the block reservation
+ * requirements as low as possible.
+ */
+STATIC int
+xfs_reflink_end_cow_extent(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		*offset_fsb,
+	xfs_fileoff_t		end_fsb)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	unsigned int		resblks;
+	int			error;
+
+	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
+			XFS_TRANS_RESERVE, &tp);
+	if (error)
+		return error;
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, ip, 0);
+
+	error = xfs_reflink_end_cow_extent_locked(tp, ip, offset_fsb, end_fsb);
+	if (error)
+		xfs_trans_cancel(tp);
+	else
+		error = xfs_trans_commit(tp);
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;
 }
@@ -761,11 +945,11 @@ xfs_reflink_end_cow(
 	end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
 
 	/*
-	 * Walk backwards until we're out of the I/O range.  The loop function
+	 * Walk forwards until we've remapped the I/O range.  The loop function
 	 * repeatedly cycles the ILOCK to allocate one transaction per remapped
 	 * extent.
 	 *
-	 * If we're being called by writeback then the the pages will still
+	 * If we're being called by writeback then the pages will still
 	 * have PageWriteback set, which prevents races with reflink remapping
 	 * and truncate.  Reflink remapping prevents races with writeback by
 	 * taking the iolock and mmaplock before flushing the pages and
@@ -793,7 +977,7 @@ xfs_reflink_end_cow(
 	 * blocks will be remapped.
 	 */
 	while (end_fsb > offset_fsb && !error)
-		error = xfs_reflink_end_cow_extent(ip, offset_fsb, &end_fsb);
+		error = xfs_reflink_end_cow_extent(ip, &offset_fsb, end_fsb);
 
 	if (error)
 		trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
@@ -801,25 +985,111 @@ xfs_reflink_end_cow(
 }
 
 /*
- * Free leftover CoW reservations that didn't get cleaned out.
+ * Fully remap all of the file's data fork at once, which is the critical part
+ * in achieving atomic behaviour.
+ * The regular CoW end path does not use function as to keep the block
+ * reservation per transaction as low as possible.
+ */
+int
+xfs_reflink_end_atomic_cow(
+	struct xfs_inode		*ip,
+	xfs_off_t			offset,
+	xfs_off_t			count)
+{
+	xfs_fileoff_t			offset_fsb;
+	xfs_fileoff_t			end_fsb;
+	int				error = 0;
+	struct xfs_mount		*mp = ip->i_mount;
+	struct xfs_trans		*tp;
+	unsigned int			resblks;
+
+	trace_xfs_reflink_end_cow(ip, offset, count);
+
+	offset_fsb = XFS_B_TO_FSBT(mp, offset);
+	end_fsb = XFS_B_TO_FSB(mp, offset + count);
+
+	/*
+	 * Each remapping operation could cause a btree split, so in the worst
+	 * case that's one for each block.
+	 */
+	resblks = (end_fsb - offset_fsb) *
+			XFS_NEXTENTADD_SPACE_RES(mp, 1, XFS_DATA_FORK);
+
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_atomic_ioend, resblks, 0,
+			XFS_TRANS_RESERVE, &tp);
+	if (error)
+		return error;
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, ip, 0);
+
+	while (end_fsb > offset_fsb && !error) {
+		error = xfs_reflink_end_cow_extent_locked(tp, ip, &offset_fsb,
+				end_fsb);
+	}
+	if (error) {
+		trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
+		goto out_cancel;
+	}
+	error = xfs_trans_commit(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	return error;
+out_cancel:
+	xfs_trans_cancel(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	return error;
+}
+
+/* Compute the largest atomic write that we can complete through software. */
+xfs_extlen_t
+xfs_reflink_max_atomic_cow(
+	struct xfs_mount	*mp)
+{
+	/* We cannot do any atomic writes without out of place writes. */
+	if (!xfs_can_sw_atomic_write(mp))
+		return 0;
+
+	/*
+	 * Atomic write limits must always be a power-of-2, according to
+	 * generic_atomic_write_valid.
+	 */
+	return rounddown_pow_of_two(xfs_calc_max_atomic_write_fsblocks(mp));
+}
+
+/*
+ * Free all CoW staging blocks that are still referenced by the ondisk refcount
+ * metadata.  The ondisk metadata does not track which inode created the
+ * staging extent, so callers must ensure that there are no cached inodes with
+ * live CoW staging extents.
  */
 int
 xfs_reflink_recover_cow(
 	struct xfs_mount	*mp)
 {
-	xfs_agnumber_t		agno;
+	struct xfs_perag	*pag = NULL;
+	struct xfs_rtgroup	*rtg = NULL;
 	int			error = 0;
 
-	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+	if (!xfs_has_reflink(mp))
 		return 0;
 
-	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
-		error = xfs_refcount_recover_cow_leftovers(mp, agno);
-		if (error)
-			break;
+	while ((pag = xfs_perag_next(mp, pag))) {
+		error = xfs_refcount_recover_cow_leftovers(pag_group(pag));
+		if (error) {
+			xfs_perag_rele(pag);
+			return error;
+		}
 	}
 
-	return error;
+	while ((rtg = xfs_rtgroup_next(mp, rtg))) {
+		error = xfs_refcount_recover_cow_leftovers(rtg_group(rtg));
+		if (error) {
+			xfs_rtgroup_rele(rtg);
+			return error;
+		}
+	}
+
+	return 0;
 }
 
 /*
@@ -926,7 +1196,7 @@ xfs_reflink_set_inode_flag(
 	if (!xfs_is_reflink_inode(src)) {
 		trace_xfs_reflink_set_inode_flag(src);
 		xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
-		src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
+		src->i_diflags2 |= XFS_DIFLAG2_REFLINK;
 		xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
 		xfs_ifork_init_cow(src);
 	} else
@@ -938,7 +1208,7 @@ xfs_reflink_set_inode_flag(
 	if (!xfs_is_reflink_inode(dest)) {
 		trace_xfs_reflink_set_inode_flag(dest);
 		xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
-		dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
+		dest->i_diflags2 |= XFS_DIFLAG2_REFLINK;
 		xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
 		xfs_ifork_init_cow(dest);
 	} else
@@ -982,12 +1252,12 @@ xfs_reflink_update_dest(
 	if (newlen > i_size_read(VFS_I(dest))) {
 		trace_xfs_reflink_update_inode_size(dest, newlen);
 		i_size_write(VFS_I(dest), newlen);
-		dest->i_d.di_size = newlen;
+		dest->i_disk_size = newlen;
 	}
 
 	if (cowextsize) {
-		dest->i_d.di_cowextsize = cowextsize;
-		dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
+		dest->i_cowextsize = cowextsize;
+		dest->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
 	}
 
 	xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
@@ -1011,14 +1281,22 @@ out_error:
 static int
 xfs_reflink_ag_has_free_space(
 	struct xfs_mount	*mp,
-	xfs_agnumber_t		agno)
+	struct xfs_inode	*ip,
+	xfs_fsblock_t		fsb)
 {
 	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
 	int			error = 0;
 
-	if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
+	if (!xfs_has_rmapbt(mp))
 		return 0;
+	if (XFS_IS_REALTIME_INODE(ip)) {
+		if (xfs_metafile_resv_critical(mp))
+			return -ENOSPC;
+		return 0;
+	}
 
+	agno = XFS_FSB_TO_AGNO(mp, fsb);
 	pag = xfs_perag_get(mp, agno);
 	if (xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) ||
 	    xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
@@ -1028,131 +1306,223 @@ xfs_reflink_ag_has_free_space(
 }
 
 /*
- * Unmap a range of blocks from a file, then map other blocks into the hole.
- * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
- * The extent irec is mapped into dest at irec->br_startoff.
+ * Remap the given extent into the file.  The dmap blockcount will be set to
+ * the number of blocks that were actually remapped.
  */
 STATIC int
 xfs_reflink_remap_extent(
 	struct xfs_inode	*ip,
-	struct xfs_bmbt_irec	*irec,
-	xfs_fileoff_t		destoff,
+	struct xfs_bmbt_irec	*dmap,
 	xfs_off_t		new_isize)
 {
+	struct xfs_bmbt_irec	smap;
 	struct xfs_mount	*mp = ip->i_mount;
-	bool			real_extent = xfs_bmap_is_real_extent(irec);
 	struct xfs_trans	*tp;
-	unsigned int		resblks;
-	struct xfs_bmbt_irec	uirec;
-	xfs_filblks_t		rlen;
-	xfs_filblks_t		unmap_len;
 	xfs_off_t		newlen;
+	int64_t			qdelta = 0;
+	unsigned int		dblocks, rblocks, resblks;
+	bool			quota_reserved = true;
+	bool			smap_real;
+	bool			dmap_written = xfs_bmap_is_written_extent(dmap);
+	bool			isrt = XFS_IS_REALTIME_INODE(ip);
+	int			iext_delta = 0;
+	int			nimaps;
 	int			error;
 
-	unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
-	trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
-
-	/* No reflinking if we're low on space */
-	if (real_extent) {
-		error = xfs_reflink_ag_has_free_space(mp,
-				XFS_FSB_TO_AGNO(mp, irec->br_startblock));
-		if (error)
-			goto out;
+	/*
+	 * Start a rolling transaction to switch the mappings.
+	 *
+	 * Adding a written extent to the extent map can cause a bmbt split,
+	 * and removing a mapped extent from the extent can cause a bmbt split.
+	 * The two operations cannot both cause a split since they operate on
+	 * the same index in the bmap btree, so we only need a reservation for
+	 * one bmbt split if either thing is happening.  However, we haven't
+	 * locked the inode yet, so we reserve assuming this is the case.
+	 *
+	 * The first allocation call tries to reserve enough space to handle
+	 * mapping dmap into a sparse part of the file plus the bmbt split.  We
+	 * haven't locked the inode or read the existing mapping yet, so we do
+	 * not know for sure that we need the space.  This should succeed most
+	 * of the time.
+	 *
+	 * If the first attempt fails, try again but reserving only enough
+	 * space to handle a bmbt split.  This is the hard minimum requirement,
+	 * and we revisit quota reservations later when we know more about what
+	 * we're remapping.
+	 */
+	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
+	if (XFS_IS_REALTIME_INODE(ip)) {
+		dblocks = resblks;
+		rblocks = dmap->br_blockcount;
+	} else {
+		dblocks = resblks + dmap->br_blockcount;
+		rblocks = 0;
+	}
+	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
+			dblocks, rblocks, false, &tp);
+	if (error == -EDQUOT || error == -ENOSPC) {
+		quota_reserved = false;
+		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
+				resblks, 0, false, &tp);
 	}
-
-	/* Start a rolling transaction to switch the mappings */
-	resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
-	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
 	if (error)
 		goto out;
 
-	xfs_ilock(ip, XFS_ILOCK_EXCL);
-	xfs_trans_ijoin(tp, ip, 0);
+	/*
+	 * Read what's currently mapped in the destination file into smap.
+	 * If smap isn't a hole, we will have to remove it before we can add
+	 * dmap to the destination file.
+	 */
+	nimaps = 1;
+	error = xfs_bmapi_read(ip, dmap->br_startoff, dmap->br_blockcount,
+			&smap, &nimaps, 0);
+	if (error)
+		goto out_cancel;
+	ASSERT(nimaps == 1 && smap.br_startoff == dmap->br_startoff);
+	smap_real = xfs_bmap_is_real_extent(&smap);
+
+	/*
+	 * We can only remap as many blocks as the smaller of the two extent
+	 * maps, because we can only remap one extent at a time.
+	 */
+	dmap->br_blockcount = min(dmap->br_blockcount, smap.br_blockcount);
+	ASSERT(dmap->br_blockcount == smap.br_blockcount);
+
+	trace_xfs_reflink_remap_extent_dest(ip, &smap);
+
+	/*
+	 * Two extents mapped to the same physical block must not have
+	 * different states; that's filesystem corruption.  Move on to the next
+	 * extent if they're both holes or both the same physical extent.
+	 */
+	if (dmap->br_startblock == smap.br_startblock) {
+		if (dmap->br_state != smap.br_state) {
+			xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
+			error = -EFSCORRUPTED;
+		}
+		goto out_cancel;
+	}
 
-	/* If we're not just clearing space, then do we have enough quota? */
-	if (real_extent) {
-		error = xfs_trans_reserve_quota_nblks(tp, ip,
-				irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS);
+	/* If both extents are unwritten, leave them alone. */
+	if (dmap->br_state == XFS_EXT_UNWRITTEN &&
+	    smap.br_state == XFS_EXT_UNWRITTEN)
+		goto out_cancel;
+
+	/* No reflinking if the AG of the dest mapping is low on space. */
+	if (dmap_written) {
+		error = xfs_reflink_ag_has_free_space(mp, ip,
+				dmap->br_startblock);
 		if (error)
 			goto out_cancel;
 	}
 
-	trace_xfs_reflink_remap(ip, irec->br_startoff,
-				irec->br_blockcount, irec->br_startblock);
-
-	/* Unmap the old blocks in the data fork. */
-	rlen = unmap_len;
-	while (rlen) {
-		ASSERT(tp->t_firstblock == NULLFSBLOCK);
-		error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1);
+	/*
+	 * Increase quota reservation if we think the quota block counter for
+	 * this file could increase.
+	 *
+	 * If we are mapping a written extent into the file, we need to have
+	 * enough quota block count reservation to handle the blocks in that
+	 * extent.  We log only the delta to the quota block counts, so if the
+	 * extent we're unmapping also has blocks allocated to it, we don't
+	 * need a quota reservation for the extent itself.
+	 *
+	 * Note that if we're replacing a delalloc reservation with a written
+	 * extent, we have to take the full quota reservation because removing
+	 * the delalloc reservation gives the block count back to the quota
+	 * count.  This is suboptimal, but the VFS flushed the dest range
+	 * before we started.  That should have removed all the delalloc
+	 * reservations, but we code defensively.
+	 *
+	 * xfs_trans_alloc_inode above already tried to grab an even larger
+	 * quota reservation, and kicked off a blockgc scan if it couldn't.
+	 * If we can't get a potentially smaller quota reservation now, we're
+	 * done.
+	 */
+	if (!quota_reserved && !smap_real && dmap_written) {
+		if (XFS_IS_REALTIME_INODE(ip)) {
+			dblocks = 0;
+			rblocks = dmap->br_blockcount;
+		} else {
+			dblocks = dmap->br_blockcount;
+			rblocks = 0;
+		}
+		error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks,
+				false);
 		if (error)
 			goto out_cancel;
+	}
 
-		/*
-		 * Trim the extent to whatever got unmapped.
-		 * Remember, bunmapi works backwards.
-		 */
-		uirec.br_startblock = irec->br_startblock + rlen;
-		uirec.br_startoff = irec->br_startoff + rlen;
-		uirec.br_blockcount = unmap_len - rlen;
-		unmap_len = rlen;
+	if (smap_real)
+		++iext_delta;
 
-		/* If this isn't a real mapping, we're done. */
-		if (!real_extent || uirec.br_blockcount == 0)
-			goto next_extent;
+	if (dmap_written)
+		++iext_delta;
 
-		trace_xfs_reflink_remap(ip, uirec.br_startoff,
-				uirec.br_blockcount, uirec.br_startblock);
+	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK, iext_delta);
+	if (error)
+		goto out_cancel;
 
-		/* Update the refcount tree */
-		error = xfs_refcount_increase_extent(tp, &uirec);
-		if (error)
-			goto out_cancel;
+	if (smap_real) {
+		/*
+		 * If the extent we're unmapping is backed by storage (written
+		 * or not), unmap the extent and drop its refcount.
+		 */
+		xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &smap);
+		xfs_refcount_decrease_extent(tp, isrt, &smap);
+		qdelta -= smap.br_blockcount;
+	} else if (smap.br_startblock == DELAYSTARTBLOCK) {
+		int		done;
 
-		/* Map the new blocks into the data fork. */
-		error = xfs_bmap_map_extent(tp, ip, &uirec);
+		/*
+		 * If the extent we're unmapping is a delalloc reservation,
+		 * we can use the regular bunmapi function to release the
+		 * incore state.  Dropping the delalloc reservation takes care
+		 * of the quota reservation for us.
+		 */
+		error = xfs_bunmapi(NULL, ip, smap.br_startoff,
+				smap.br_blockcount, 0, 1, &done);
 		if (error)
 			goto out_cancel;
+		ASSERT(done);
+	}
 
-		/* Update quota accounting. */
-		xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
-				uirec.br_blockcount);
-
-		/* Update dest isize if needed. */
-		newlen = XFS_FSB_TO_B(mp,
-				uirec.br_startoff + uirec.br_blockcount);
-		newlen = min_t(xfs_off_t, newlen, new_isize);
-		if (newlen > i_size_read(VFS_I(ip))) {
-			trace_xfs_reflink_update_inode_size(ip, newlen);
-			i_size_write(VFS_I(ip), newlen);
-			ip->i_d.di_size = newlen;
-			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
-		}
+	/*
+	 * If the extent we're sharing is backed by written storage, increase
+	 * its refcount and map it into the file.
+	 */
+	if (dmap_written) {
+		xfs_refcount_increase_extent(tp, isrt, dmap);
+		xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, dmap);
+		qdelta += dmap->br_blockcount;
+	}
 
-next_extent:
-		/* Process all the deferred stuff. */
-		error = xfs_defer_finish(&tp);
-		if (error)
-			goto out_cancel;
+	xfs_reflink_update_quota(tp, ip, false, qdelta);
+
+	/* Update dest isize if needed. */
+	newlen = XFS_FSB_TO_B(mp, dmap->br_startoff + dmap->br_blockcount);
+	newlen = min_t(xfs_off_t, newlen, new_isize);
+	if (newlen > i_size_read(VFS_I(ip))) {
+		trace_xfs_reflink_update_inode_size(ip, newlen);
+		i_size_write(VFS_I(ip), newlen);
+		ip->i_disk_size = newlen;
+		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 	}
 
+	/* Commit everything and unlock. */
 	error = xfs_trans_commit(tp);
-	xfs_iunlock(ip, XFS_ILOCK_EXCL);
-	if (error)
-		goto out;
-	return 0;
+	goto out_unlock;
 
 out_cancel:
 	xfs_trans_cancel(tp);
+out_unlock:
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 out:
-	trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
+	if (error)
+		trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
 	return error;
 }
 
-/*
- * Iteratively remap one file's extents (and holes) to another's.
- */
+/* Remap a range of one file to the other. */
 int
 xfs_reflink_remap_blocks(
 	struct xfs_inode	*src,
@@ -1163,25 +1533,22 @@ xfs_reflink_remap_blocks(
 	loff_t			*remapped)
 {
 	struct xfs_bmbt_irec	imap;
-	xfs_fileoff_t		srcoff;
-	xfs_fileoff_t		destoff;
+	struct xfs_mount	*mp = src->i_mount;
+	xfs_fileoff_t		srcoff = XFS_B_TO_FSBT(mp, pos_in);
+	xfs_fileoff_t		destoff = XFS_B_TO_FSBT(mp, pos_out);
 	xfs_filblks_t		len;
-	xfs_filblks_t		range_len;
 	xfs_filblks_t		remapped_len = 0;
 	xfs_off_t		new_isize = pos_out + remap_len;
 	int			nimaps;
 	int			error = 0;
 
-	destoff = XFS_B_TO_FSBT(src->i_mount, pos_out);
-	srcoff = XFS_B_TO_FSBT(src->i_mount, pos_in);
-	len = XFS_B_TO_FSB(src->i_mount, remap_len);
+	len = min_t(xfs_filblks_t, XFS_B_TO_FSB(mp, remap_len),
+			XFS_MAX_FILEOFF);
 
-	/* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
-	while (len) {
-		uint		lock_mode;
+	trace_xfs_reflink_remap_blocks(src, srcoff, len, dest, destoff);
 
-		trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
-				dest, destoff);
+	while (len > 0) {
+		unsigned int	lock_mode;
 
 		/* Read extent from the source file */
 		nimaps = 1;
@@ -1190,18 +1557,26 @@ xfs_reflink_remap_blocks(
 		xfs_iunlock(src, lock_mode);
 		if (error)
 			break;
-		ASSERT(nimaps == 1);
-
-		trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
-				&imap);
+		/*
+		 * The caller supposedly flushed all dirty pages in the source
+		 * file range, which means that writeback should have allocated
+		 * or deleted all delalloc reservations in that range.  If we
+		 * find one, that's a good sign that something is seriously
+		 * wrong here.
+		 */
+		ASSERT(nimaps == 1 && imap.br_startoff == srcoff);
+		if (imap.br_startblock == DELAYSTARTBLOCK) {
+			ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
+			xfs_bmap_mark_sick(src, XFS_DATA_FORK);
+			error = -EFSCORRUPTED;
+			break;
+		}
 
-		/* Translate imap into the destination file. */
-		range_len = imap.br_startoff + imap.br_blockcount - srcoff;
-		imap.br_startoff += destoff - srcoff;
+		trace_xfs_reflink_remap_extent_src(src, &imap);
 
-		/* Clear dest from destoff to the end of imap and map it in. */
-		error = xfs_reflink_remap_extent(dest, &imap, destoff,
-				new_isize);
+		/* Remap into the destination file at the given offset. */
+		imap.br_startoff = destoff;
+		error = xfs_reflink_remap_extent(dest, &imap, new_isize);
 		if (error)
 			break;
 
@@ -1211,10 +1586,11 @@ xfs_reflink_remap_blocks(
 		}
 
 		/* Advance drange/srange */
-		srcoff += range_len;
-		destoff += range_len;
-		len -= range_len;
-		remapped_len += range_len;
+		srcoff += imap.br_blockcount;
+		destoff += imap.br_blockcount;
+		len -= imap.br_blockcount;
+		remapped_len += imap.br_blockcount;
+		cond_resched();
 	}
 
 	if (error)
@@ -1225,70 +1601,6 @@ xfs_reflink_remap_blocks(
 }
 
 /*
- * Grab the exclusive iolock for a data copy from src to dest, making
- * sure to abide vfs locking order (lowest pointer value goes first) and
- * breaking the pnfs layout leases on dest before proceeding.  The loop
- * is needed because we cannot call the blocking break_layout() with the
- * src iolock held, and therefore have to back out both locks.
- */
-static int
-xfs_iolock_two_inodes_and_break_layout(
-	struct inode		*src,
-	struct inode		*dest)
-{
-	int			error;
-
-retry:
-	if (src < dest) {
-		inode_lock_shared(src);
-		inode_lock_nested(dest, I_MUTEX_NONDIR2);
-	} else {
-		/* src >= dest */
-		inode_lock(dest);
-	}
-
-	error = break_layout(dest, false);
-	if (error == -EWOULDBLOCK) {
-		inode_unlock(dest);
-		if (src < dest)
-			inode_unlock_shared(src);
-		error = break_layout(dest, true);
-		if (error)
-			return error;
-		goto retry;
-	}
-	if (error) {
-		inode_unlock(dest);
-		if (src < dest)
-			inode_unlock_shared(src);
-		return error;
-	}
-	if (src > dest)
-		inode_lock_shared_nested(src, I_MUTEX_NONDIR2);
-	return 0;
-}
-
-/* Unlock both inodes after they've been prepped for a range clone. */
-void
-xfs_reflink_remap_unlock(
-	struct file		*file_in,
-	struct file		*file_out)
-{
-	struct inode		*inode_in = file_inode(file_in);
-	struct xfs_inode	*src = XFS_I(inode_in);
-	struct inode		*inode_out = file_inode(file_out);
-	struct xfs_inode	*dest = XFS_I(inode_out);
-	bool			same_inode = (inode_in == inode_out);
-
-	xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
-	if (!same_inode)
-		xfs_iunlock(src, XFS_MMAPLOCK_SHARED);
-	inode_unlock(inode_out);
-	if (!same_inode)
-		inode_unlock_shared(inode_in);
-}
-
-/*
  * If we're reflinking to a point past the destination file's EOF, we must
  * zero any speculative post-EOF preallocations that sit between the old EOF
  * and the destination file offset.
@@ -1304,8 +1616,7 @@ xfs_reflink_zero_posteof(
 		return 0;
 
 	trace_xfs_zero_eof(ip, isize, pos - isize);
-	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
-			&xfs_iomap_ops);
+	return xfs_zero_range(ip, isize, pos - isize, NULL, NULL);
 }
 
 /*
@@ -1350,32 +1661,30 @@ xfs_reflink_remap_prep(
 	struct xfs_inode	*src = XFS_I(inode_in);
 	struct inode		*inode_out = file_inode(file_out);
 	struct xfs_inode	*dest = XFS_I(inode_out);
-	bool			same_inode = (inode_in == inode_out);
-	ssize_t			ret;
+	int			ret;
 
 	/* Lock both files against IO */
-	ret = xfs_iolock_two_inodes_and_break_layout(inode_in, inode_out);
+	ret = xfs_ilock2_io_mmap(src, dest);
 	if (ret)
 		return ret;
-	if (same_inode)
-		xfs_ilock(src, XFS_MMAPLOCK_EXCL);
-	else
-		xfs_lock_two_inodes(src, XFS_MMAPLOCK_SHARED, dest,
-				XFS_MMAPLOCK_EXCL);
 
 	/* Check file eligibility and prepare for block sharing. */
 	ret = -EINVAL;
-	/* Don't reflink realtime inodes */
-	if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
+	/* Can't reflink between data and rt volumes */
+	if (XFS_IS_REALTIME_INODE(src) != XFS_IS_REALTIME_INODE(dest))
 		goto out_unlock;
 
-	/* Don't share DAX file data for now. */
-	if (IS_DAX(inode_in) || IS_DAX(inode_out))
+	/* Don't share DAX file data with non-DAX file. */
+	if (IS_DAX(inode_in) != IS_DAX(inode_out))
 		goto out_unlock;
 
-	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
-			len, remap_flags);
-	if (ret < 0 || *len == 0)
+	if (!IS_DAX(inode_in))
+		ret = generic_remap_file_range_prep(file_in, pos_in, file_out,
+				pos_out, len, remap_flags);
+	else
+		ret = dax_remap_file_range_prep(file_in, pos_in, file_out,
+				pos_out, len, remap_flags, &xfs_read_iomap_ops);
+	if (ret || *len == 0)
 		goto out_unlock;
 
 	/* Attach dquots to dest inode before changing block map */
@@ -1410,91 +1719,16 @@ xfs_reflink_remap_prep(
 	if (ret)
 		goto out_unlock;
 
-	return 1;
+	xfs_iflags_set(src, XFS_IREMAPPING);
+	if (inode_in != inode_out)
+		xfs_ilock_demote(src, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
+
+	return 0;
 out_unlock:
-	xfs_reflink_remap_unlock(file_in, file_out);
+	xfs_iunlock2_io_mmap(src, dest);
 	return ret;
 }
 
-/*
- * The user wants to preemptively CoW all shared blocks in this file,
- * which enables us to turn off the reflink flag.  Iterate all
- * extents which are not prealloc/delalloc to see which ranges are
- * mentioned in the refcount tree, then read those blocks into the
- * pagecache, dirty them, fsync them back out, and then we can update
- * the inode flag.  What happens if we run out of memory? :)
- */
-STATIC int
-xfs_reflink_dirty_extents(
-	struct xfs_inode	*ip,
-	xfs_fileoff_t		fbno,
-	xfs_filblks_t		end,
-	xfs_off_t		isize)
-{
-	struct xfs_mount	*mp = ip->i_mount;
-	xfs_agnumber_t		agno;
-	xfs_agblock_t		agbno;
-	xfs_extlen_t		aglen;
-	xfs_agblock_t		rbno;
-	xfs_extlen_t		rlen;
-	xfs_off_t		fpos;
-	xfs_off_t		flen;
-	struct xfs_bmbt_irec	map[2];
-	int			nmaps;
-	int			error = 0;
-
-	while (end - fbno > 0) {
-		nmaps = 1;
-		/*
-		 * Look for extents in the file.  Skip holes, delalloc, or
-		 * unwritten extents; they can't be reflinked.
-		 */
-		error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0);
-		if (error)
-			goto out;
-		if (nmaps == 0)
-			break;
-		if (!xfs_bmap_is_real_extent(&map[0]))
-			goto next;
-
-		map[1] = map[0];
-		while (map[1].br_blockcount) {
-			agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock);
-			agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock);
-			aglen = map[1].br_blockcount;
-
-			error = xfs_reflink_find_shared(mp, NULL, agno, agbno,
-					aglen, &rbno, &rlen, true);
-			if (error)
-				goto out;
-			if (rbno == NULLAGBLOCK)
-				break;
-
-			/* Dirty the pages */
-			xfs_iunlock(ip, XFS_ILOCK_EXCL);
-			fpos = XFS_FSB_TO_B(mp, map[1].br_startoff +
-					(rbno - agbno));
-			flen = XFS_FSB_TO_B(mp, rlen);
-			if (fpos + flen > isize)
-				flen = isize - fpos;
-			error = iomap_file_dirty(VFS_I(ip), fpos, flen,
-					&xfs_iomap_ops);
-			xfs_ilock(ip, XFS_ILOCK_EXCL);
-			if (error)
-				goto out;
-
-			map[1].br_blockcount -= (rbno - agbno + rlen);
-			map[1].br_startoff += (rbno - agbno + rlen);
-			map[1].br_startblock += (rbno - agbno + rlen);
-		}
-
-next:
-		fbno = map[0].br_startoff + map[0].br_blockcount;
-	}
-out:
-	return error;
-}
-
 /* Does this inode need the reflink flag? */
 int
 xfs_reflink_inode_has_shared_extents(
@@ -1505,38 +1739,35 @@ xfs_reflink_inode_has_shared_extents(
 	struct xfs_bmbt_irec		got;
 	struct xfs_mount		*mp = ip->i_mount;
 	struct xfs_ifork		*ifp;
-	xfs_agnumber_t			agno;
-	xfs_agblock_t			agbno;
-	xfs_extlen_t			aglen;
-	xfs_agblock_t			rbno;
-	xfs_extlen_t			rlen;
 	struct xfs_iext_cursor		icur;
 	bool				found;
 	int				error;
 
-	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
-	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
-		error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
-		if (error)
-			return error;
-	}
+	ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
+	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
+	if (error)
+		return error;
 
 	*has_shared = false;
 	found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got);
 	while (found) {
+		xfs_extlen_t		shared_offset, shared_len;
+
 		if (isnullstartblock(got.br_startblock) ||
 		    got.br_state != XFS_EXT_NORM)
 			goto next;
-		agno = XFS_FSB_TO_AGNO(mp, got.br_startblock);
-		agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock);
-		aglen = got.br_blockcount;
 
-		error = xfs_reflink_find_shared(mp, tp, agno, agbno, aglen,
-				&rbno, &rlen, false);
+		if (XFS_IS_REALTIME_INODE(ip))
+			error = xfs_reflink_find_rtshared(mp, tp, &got,
+					&shared_offset, &shared_len, false);
+		else
+			error = xfs_reflink_find_shared(mp, tp, &got,
+					&shared_offset, &shared_len, false);
 		if (error)
 			return error;
+
 		/* Is there still a shared block here? */
-		if (rbno != NULLAGBLOCK) {
+		if (shared_len) {
 			*has_shared = true;
 			return 0;
 		}
@@ -1563,6 +1794,9 @@ xfs_reflink_clear_inode_flag(
 
 	ASSERT(xfs_is_reflink_inode(ip));
 
+	if (!xfs_can_free_cowblocks(ip))
+		return 0;
+
 	error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag);
 	if (error || needs_flag)
 		return error;
@@ -1571,13 +1805,14 @@ xfs_reflink_clear_inode_flag(
 	 * We didn't find any shared blocks so turn off the reflink flag.
 	 * First, get rid of any leftover CoW mappings.
 	 */
-	error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true);
+	error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, XFS_MAX_FILEOFF,
+			true);
 	if (error)
 		return error;
 
 	/* Clear the inode flag. */
 	trace_xfs_reflink_unset_inode_flag(ip);
-	ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+	ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
 	xfs_inode_clear_cowblocks_tag(ip);
 	xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
 
@@ -1631,10 +1866,7 @@ xfs_reflink_unshare(
 	xfs_off_t		offset,
 	xfs_off_t		len)
 {
-	struct xfs_mount	*mp = ip->i_mount;
-	xfs_fileoff_t		fbno;
-	xfs_filblks_t		end;
-	xfs_off_t		isize;
+	struct inode		*inode = VFS_I(ip);
 	int			error;
 
 	if (!xfs_is_reflink_inode(ip))
@@ -1642,20 +1874,20 @@ xfs_reflink_unshare(
 
 	trace_xfs_reflink_unshare(ip, offset, len);
 
-	inode_dio_wait(VFS_I(ip));
+	inode_dio_wait(inode);
 
-	/* Try to CoW the selected ranges */
-	xfs_ilock(ip, XFS_ILOCK_EXCL);
-	fbno = XFS_B_TO_FSBT(mp, offset);
-	isize = i_size_read(VFS_I(ip));
-	end = XFS_B_TO_FSB(mp, offset + len);
-	error = xfs_reflink_dirty_extents(ip, fbno, end, isize);
+	if (IS_DAX(inode))
+		error = dax_file_unshare(inode, offset, len,
+				&xfs_dax_write_iomap_ops);
+	else
+		error = iomap_file_unshare(inode, offset, len,
+				&xfs_buffered_write_iomap_ops,
+				&xfs_iomap_write_ops);
 	if (error)
-		goto out_unlock;
-	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+		goto out;
 
-	/* Wait for the IO to finish */
-	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
+	error = filemap_write_and_wait_range(inode->i_mapping, offset,
+			offset + len - 1);
 	if (error)
 		goto out;
 
@@ -1663,12 +1895,34 @@ xfs_reflink_unshare(
 	error = xfs_reflink_try_clear_inode_flag(ip);
 	if (error)
 		goto out;
-
 	return 0;
 
-out_unlock:
-	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 out:
 	trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);
 	return error;
 }
+
+/*
+ * Can we use reflink with this realtime extent size?  Note that we don't check
+ * for rblocks > 0 here because this can be called as part of attaching a new
+ * rt section.
+ */
+bool
+xfs_reflink_supports_rextsize(
+	struct xfs_mount	*mp,
+	unsigned int		rextsize)
+{
+	/* reflink on the realtime device requires rtgroups */
+	if (!xfs_has_rtgroups(mp))
+	       return false;
+
+	/*
+	 * Reflink doesn't support rt extent size larger than a single fsblock
+	 * because we would have to perform CoW-around for unaligned write
+	 * requests to guarantee that we always remap entire rt extents.
+	 */
+	if (rextsize != 1)
+		return false;
+
+	return true;
+}