Merge tag 'xfs-reflink-for-linus-4.9-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs

    < XFS has gained super CoW powers! >
     ----------------------------------
            \   ^__^
             \  (oo)\_______
                (__)\       )\/\
                    ||----w |
                    ||     ||

Pull XFS support for shared data extents from Dave Chinner:
 "This is the second part of the XFS updates for this merge cycle.  This
  pullreq contains the new shared data extents feature for XFS.

  Given the complexity and size of this change I am expecting - like the
  addition of reverse mapping last cycle - that there will be some
  follow-up bug fixes and cleanups around the -rc3 stage for issues that
  I'm sure will show up once the code hits a wider userbase.

  What it is:

  At the most basic level we are simply adding shared data extents to
  XFS - i.e. a single extent on disk can now have multiple owners. To do
  this we have to add new on-disk features to both track the shared
  extents and the number of times they've been shared. This is done by
  the new "refcount" btree that sits in every allocation group. When we
  share or unshare an extent, this tree gets updated.

  Along with this new tree, the reverse mapping tree needs to be updated
  to track each owner or a shared extent. This also needs to be updated
  ever share/unshare operation. These interactions at extent allocation
  and freeing time have complex ordering and recovery constraints, so
  there's a significant amount of new intent-based transaction code to
  ensure that operations are performed atomically from both the runtime
  and integrity/crash recovery perspectives.

  We also need to break sharing when writes hit a shared extent - this
  is where the new copy-on-write implementation comes in. We allocate
  new storage and copy the original data along with the overwrite data
  into the new location. We only do this for data as we don't share
  metadata at all - each inode has it's own metadata that tracks the
  shared data extents, the extents undergoing CoW and it's own private
  extents.

  Of course, being XFS, nothing is simple - we use delayed allocation
  for CoW similar to how we use it for normal writes. ENOSPC is a
  significant issue here - we build on the reservation code added in
  4.8-rc1 with the reverse mapping feature to ensure we don't get
  spurious ENOSPC issues part way through a CoW operation. These
  mechanisms also help minimise fragmentation due to repeated CoW
  operations. To further reduce fragmentation overhead, we've also
  introduced a CoW extent size hint, which indicates how large a region
  we should allocate when we execute a CoW operation.

  With all this functionality in place, we can hook up .copy_file_range,
  .clone_file_range and .dedupe_file_range and we gain all the
  capabilities of reflink and other vfs provided functionality that
  enable manipulation to shared extents. We also added a fallocate mode
  that explicitly unshares a range of a file, which we implemented as an
  explicit CoW of all the shared extents in a file.

  As such, it's a huge chunk of new functionality with new on-disk
  format features and internal infrastructure. It warns at mount time as
  an experimental feature and that it may eat data (as we do with all
  new on-disk features until they stabilise). We have not released
  userspace suport for it yet - userspace support currently requires
  download from Darrick's xfsprogs repo and build from source, so the
  access to this feature is really developer/tester only at this point.
  Initial userspace support will be released at the same time the kernel
  with this code in it is released.

  The new code causes 5-6 new failures with xfstests - these aren't
  serious functional failures but things the output of tests changing
  slightly due to perturbations in layouts, space usage, etc. OTOH,
  we've added 150+ new tests to xfstests that specifically exercise this
  new functionality so it's got far better test coverage than any
  functionality we've previously added to XFS.

  Darrick has done a pretty amazing job getting us to this stage, and
  special mention also needs to go to Christoph (review, testing,
  improvements and bug fixes) and Brian (caught several intricate bugs
  during review) for the effort they've also put in.

  Summary:

   - unshare range (FALLOC_FL_UNSHARE) support for fallocate

   - copy-on-write extent size hints (FS_XFLAG_COWEXTSIZE) for fsxattr
     interface

   - shared extent support for XFS

   - copy-on-write support for shared extents

   - copy_file_range support

   - clone_file_range support (implements reflink)

   - dedupe_file_range support

   - defrag support for reverse mapping enabled filesystems"

* tag 'xfs-reflink-for-linus-4.9-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (71 commits)
  xfs: convert COW blocks to real blocks before unwritten extent conversion
  xfs: rework refcount cow recovery error handling
  xfs: clear reflink flag if setting realtime flag
  xfs: fix error initialization
  xfs: fix label inaccuracies
  xfs: remove isize check from unshare operation
  xfs: reduce stack usage of _reflink_clear_inode_flag
  xfs: check inode reflink flag before calling reflink functions
  xfs: implement swapext for rmap filesystems
  xfs: refactor swapext code
  xfs: various swapext cleanups
  xfs: recognize the reflink feature bit
  xfs: simulate per-AG reservations being critically low
  xfs: don't mix reflink and DAX mode for now
  xfs: check for invalid inode reflink flags
  xfs: set a default CoW extent size of 32 blocks
  xfs: convert unwritten status of reverse mappings for shared files
  xfs: use interval query for rmap alloc operations on shared files
  xfs: add shared rmap map/unmap/convert log item types
  xfs: increase log reservations for reflink
  ...

fs/open.c

@@ -267,6 +267,11 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 	    (mode & ~FALLOC_FL_INSERT_RANGE))
 		return -EINVAL;
 
+	/* Unshare range should only be used with allocate mode. */
+	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
+	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
+		return -EINVAL;
+
 	if (!(file->f_mode & FMODE_WRITE))
 		return -EBADF;
 

fs/xfs/Makefile

@@ -55,6 +55,8 @@ xfs-y				+= $(addprefix libxfs/, \
 				   xfs_ag_resv.o \
 				   xfs_rmap.o \
 				   xfs_rmap_btree.o \
+				   xfs_refcount.o \
+				   xfs_refcount_btree.o \
 				   xfs_sb.o \
 				   xfs_symlink_remote.o \
 				   xfs_trans_resv.o \
@@ -88,6 +90,7 @@ xfs-y				+= xfs_aops.o \
 				   xfs_message.o \
 				   xfs_mount.o \
 				   xfs_mru_cache.o \
+				   xfs_reflink.o \
 				   xfs_stats.o \
 				   xfs_super.o \
 				   xfs_symlink.o \
@@ -100,16 +103,20 @@ xfs-y				+= xfs_aops.o \
 # low-level transaction/log code
 xfs-y				+= xfs_log.o \
 				   xfs_log_cil.o \
+				   xfs_bmap_item.o \
 				   xfs_buf_item.o \
 				   xfs_extfree_item.o \
 				   xfs_icreate_item.o \
 				   xfs_inode_item.o \
+				   xfs_refcount_item.o \
 				   xfs_rmap_item.o \
 				   xfs_log_recover.o \
 				   xfs_trans_ail.o \
+				   xfs_trans_bmap.o \
 				   xfs_trans_buf.o \
 				   xfs_trans_extfree.o \
 				   xfs_trans_inode.o \
+				   xfs_trans_refcount.o \
 				   xfs_trans_rmap.o \
 
 # optional features

fs/xfs/libxfs/xfs_ag_resv.c

@@ -38,6 +38,7 @@
 #include "xfs_trans_space.h"
 #include "xfs_rmap_btree.h"
 #include "xfs_btree.h"
+#include "xfs_refcount_btree.h"
 
 /*
  * Per-AG Block Reservations
@@ -108,7 +109,9 @@ xfs_ag_resv_critical(
 	trace_xfs_ag_resv_critical(pag, type, avail);
 
 	/* Critically low if less than 10% or max btree height remains. */
-	return avail < orig / 10 || avail < XFS_BTREE_MAXLEVELS;
+	return XFS_TEST_ERROR(avail < orig / 10 || avail < XFS_BTREE_MAXLEVELS,
+			pag->pag_mount, XFS_ERRTAG_AG_RESV_CRITICAL,
+			XFS_RANDOM_AG_RESV_CRITICAL);
 }
 
 /*
@@ -228,6 +231,11 @@ xfs_ag_resv_init(
 	if (pag->pag_meta_resv.ar_asked == 0) {
 		ask = used = 0;
 
+		error = xfs_refcountbt_calc_reserves(pag->pag_mount,
+				pag->pag_agno, &ask, &used);
+		if (error)
+			goto out;
+
 		error = __xfs_ag_resv_init(pag, XFS_AG_RESV_METADATA,
 				ask, used);
 		if (error)
@@ -238,6 +246,11 @@ xfs_ag_resv_init(
 	if (pag->pag_agfl_resv.ar_asked == 0) {
 		ask = used = 0;
 
+		error = xfs_rmapbt_calc_reserves(pag->pag_mount, pag->pag_agno,
+				&ask, &used);
+		if (error)
+			goto out;
+
 		error = __xfs_ag_resv_init(pag, XFS_AG_RESV_AGFL, ask, used);
 		if (error)
 			goto out;

fs/xfs/libxfs/xfs_alloc.c

@@ -52,10 +52,23 @@ STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
 		xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
 
+unsigned int
+xfs_refc_block(
+	struct xfs_mount	*mp)
+{
+	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
+		return XFS_RMAP_BLOCK(mp) + 1;
+	if (xfs_sb_version_hasfinobt(&mp->m_sb))
+		return XFS_FIBT_BLOCK(mp) + 1;
+	return XFS_IBT_BLOCK(mp) + 1;
+}
+
 xfs_extlen_t
 xfs_prealloc_blocks(
 	struct xfs_mount	*mp)
 {
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		return xfs_refc_block(mp) + 1;
 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
 		return XFS_RMAP_BLOCK(mp) + 1;
 	if (xfs_sb_version_hasfinobt(&mp->m_sb))
@@ -115,6 +128,8 @@ xfs_alloc_ag_max_usable(
 		blocks++;		/* finobt root block */
 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
 		blocks++; 		/* rmap root block */
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		blocks++;		/* refcount root block */
 
 	return mp->m_sb.sb_agblocks - blocks;
 }
@@ -2321,6 +2336,9 @@ xfs_alloc_log_agf(
 		offsetof(xfs_agf_t, agf_btreeblks),
 		offsetof(xfs_agf_t, agf_uuid),
 		offsetof(xfs_agf_t, agf_rmap_blocks),
+		offsetof(xfs_agf_t, agf_refcount_blocks),
+		offsetof(xfs_agf_t, agf_refcount_root),
+		offsetof(xfs_agf_t, agf_refcount_level),
 		/* needed so that we don't log the whole rest of the structure: */
 		offsetof(xfs_agf_t, agf_spare64),
 		sizeof(xfs_agf_t)
@@ -2458,6 +2476,10 @@ xfs_agf_verify(
 	    be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
 		return false;
 
+	if (xfs_sb_version_hasreflink(&mp->m_sb) &&
+	    be32_to_cpu(agf->agf_refcount_level) > XFS_BTREE_MAXLEVELS)
+		return false;
+
 	return true;;
 
 }
@@ -2578,6 +2600,7 @@ xfs_alloc_read_agf(
 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
 		pag->pagf_levels[XFS_BTNUM_RMAPi] =
 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
+		pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
 		spin_lock_init(&pag->pagb_lock);
 		pag->pagb_count = 0;
 		pag->pagb_tree = RB_ROOT;

fs/xfs/libxfs/xfs_bmap.c

@@ -48,6 +48,7 @@
 #include "xfs_filestream.h"
 #include "xfs_rmap.h"
 #include "xfs_ag_resv.h"
+#include "xfs_refcount.h"
 
 
 kmem_zone_t		*xfs_bmap_free_item_zone;
@@ -140,7 +141,8 @@ xfs_bmbt_lookup_ge(
  */
 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 {
-	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
+	return whichfork != XFS_COW_FORK &&
+		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 		XFS_IFORK_NEXTENTS(ip, whichfork) >
 			XFS_IFORK_MAXEXT(ip, whichfork);
 }
@@ -150,7 +152,8 @@ static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
  */
 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 {
-	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
+	return whichfork != XFS_COW_FORK &&
+		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
 			XFS_IFORK_MAXEXT(ip, whichfork);
 }
@@ -640,6 +643,7 @@ xfs_bmap_btree_to_extents(
 
 	mp = ip->i_mount;
 	ifp = XFS_IFORK_PTR(ip, whichfork);
+	ASSERT(whichfork != XFS_COW_FORK);
 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 	rblock = ifp->if_broot;
@@ -706,6 +710,7 @@ xfs_bmap_extents_to_btree(
 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
 
 	mp = ip->i_mount;
+	ASSERT(whichfork != XFS_COW_FORK);
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 
@@ -748,6 +753,7 @@ xfs_bmap_extents_to_btree(
 		args.type = XFS_ALLOCTYPE_START_BNO;
 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 	} else if (dfops->dop_low) {
+try_another_ag:
 		args.type = XFS_ALLOCTYPE_START_BNO;
 		args.fsbno = *firstblock;
 	} else {
@@ -762,6 +768,21 @@ xfs_bmap_extents_to_btree(
 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 		return error;
 	}
+
+	/*
+	 * During a CoW operation, the allocation and bmbt updates occur in
+	 * different transactions.  The mapping code tries to put new bmbt
+	 * blocks near extents being mapped, but the only way to guarantee this
+	 * is if the alloc and the mapping happen in a single transaction that
+	 * has a block reservation.  That isn't the case here, so if we run out
+	 * of space we'll try again with another AG.
+	 */
+	if (xfs_sb_version_hasreflink(&cur->bc_mp->m_sb) &&
+	    args.fsbno == NULLFSBLOCK &&
+	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
+		dfops->dop_low = true;
+		goto try_another_ag;
+	}
 	/*
 	 * Allocation can't fail, the space was reserved.
 	 */
@@ -837,6 +858,7 @@ xfs_bmap_local_to_extents_empty(
 {
 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
 
+	ASSERT(whichfork != XFS_COW_FORK);
 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 	ASSERT(ifp->if_bytes == 0);
 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
@@ -896,6 +918,7 @@ xfs_bmap_local_to_extents(
 	 * file currently fits in an inode.
 	 */
 	if (*firstblock == NULLFSBLOCK) {
+try_another_ag:
 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 		args.type = XFS_ALLOCTYPE_START_BNO;
 	} else {
@@ -908,6 +931,19 @@ xfs_bmap_local_to_extents(
 	if (error)
 		goto done;
 
+	/*
+	 * During a CoW operation, the allocation and bmbt updates occur in
+	 * different transactions.  The mapping code tries to put new bmbt
+	 * blocks near extents being mapped, but the only way to guarantee this
+	 * is if the alloc and the mapping happen in a single transaction that
+	 * has a block reservation.  That isn't the case here, so if we run out
+	 * of space we'll try again with another AG.
+	 */
+	if (xfs_sb_version_hasreflink(&ip->i_mount->m_sb) &&
+	    args.fsbno == NULLFSBLOCK &&
+	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
+		goto try_another_ag;
+	}
 	/* Can't fail, the space was reserved. */
 	ASSERT(args.fsbno != NULLFSBLOCK);
 	ASSERT(args.len == 1);
@@ -1670,7 +1706,8 @@ xfs_bmap_one_block(
  */
 STATIC int				/* error */
 xfs_bmap_add_extent_delay_real(
-	struct xfs_bmalloca	*bma)
+	struct xfs_bmalloca	*bma,
+	int			whichfork)
 {
 	struct xfs_bmbt_irec	*new = &bma->got;
 	int			diff;	/* temp value */
@@ -1688,11 +1725,14 @@ xfs_bmap_add_extent_delay_real(
 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
 	xfs_filblks_t		temp2=0;/* value for da_new calculations */
 	int			tmp_rval;	/* partial logging flags */
-	int			whichfork = XFS_DATA_FORK;
 	struct xfs_mount	*mp;
+	xfs_extnum_t		*nextents;
 
 	mp = bma->ip->i_mount;
 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
+	ASSERT(whichfork != XFS_ATTR_FORK);
+	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
+						&bma->ip->i_d.di_nextents);
 
 	ASSERT(bma->idx >= 0);
 	ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
@@ -1706,6 +1746,9 @@ xfs_bmap_add_extent_delay_real(
 #define	RIGHT		r[1]
 #define	PREV		r[2]
 
+	if (whichfork == XFS_COW_FORK)
+		state |= BMAP_COWFORK;
+
 	/*
 	 * Set up a bunch of variables to make the tests simpler.
 	 */
@@ -1792,7 +1835,7 @@ xfs_bmap_add_extent_delay_real(
 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 
 		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
-		bma->ip->i_d.di_nextents--;
+		(*nextents)--;
 		if (bma->cur == NULL)
 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 		else {
@@ -1894,7 +1937,7 @@ xfs_bmap_add_extent_delay_real(
 		xfs_bmbt_set_startblock(ep, new->br_startblock);
 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 
-		bma->ip->i_d.di_nextents++;
+		(*nextents)++;
 		if (bma->cur == NULL)
 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 		else {
@@ -1964,7 +2007,7 @@ xfs_bmap_add_extent_delay_real(
 		temp = PREV.br_blockcount - new->br_blockcount;
 		xfs_bmbt_set_blockcount(ep, temp);
 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
-		bma->ip->i_d.di_nextents++;
+		(*nextents)++;
 		if (bma->cur == NULL)
 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 		else {
@@ -2048,7 +2091,7 @@ xfs_bmap_add_extent_delay_real(
 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 		xfs_bmbt_set_blockcount(ep, temp);
 		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
-		bma->ip->i_d.di_nextents++;
+		(*nextents)++;
 		if (bma->cur == NULL)
 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 		else {
@@ -2117,7 +2160,7 @@ xfs_bmap_add_extent_delay_real(
 		RIGHT.br_blockcount = temp2;
 		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
-		bma->ip->i_d.di_nextents++;
+		(*nextents)++;
 		if (bma->cur == NULL)
 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 		else {
@@ -2215,7 +2258,8 @@ xfs_bmap_add_extent_delay_real(
 
 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
 done:
-	bma->logflags |= rval;
+	if (whichfork != XFS_COW_FORK)
+		bma->logflags |= rval;
 	return error;
 #undef	LEFT
 #undef	RIGHT
@@ -2759,6 +2803,7 @@ done:
 STATIC void
 xfs_bmap_add_extent_hole_delay(
 	xfs_inode_t		*ip,	/* incore inode pointer */
+	int			whichfork,
 	xfs_extnum_t		*idx,	/* extent number to update/insert */
 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
 {
@@ -2770,8 +2815,10 @@ xfs_bmap_add_extent_hole_delay(
 	int			state;  /* state bits, accessed thru macros */
 	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
 
-	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
+	ifp = XFS_IFORK_PTR(ip, whichfork);
 	state = 0;
+	if (whichfork == XFS_COW_FORK)
+		state |= BMAP_COWFORK;
 	ASSERT(isnullstartblock(new->br_startblock));
 
 	/*
@@ -2789,7 +2836,7 @@ xfs_bmap_add_extent_hole_delay(
 	 * Check and set flags if the current (right) segment exists.
 	 * If it doesn't exist, we're converting the hole at end-of-file.
 	 */
-	if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
+	if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
 		state |= BMAP_RIGHT_VALID;
 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
 
@@ -2923,6 +2970,7 @@ xfs_bmap_add_extent_hole_real(
 	ASSERT(!isnullstartblock(new->br_startblock));
 	ASSERT(!bma->cur ||
 	       !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
+	ASSERT(whichfork != XFS_COW_FORK);
 
 	XFS_STATS_INC(mp, xs_add_exlist);
 
@@ -3648,7 +3696,9 @@ xfs_bmap_btalloc(
 	else if (mp->m_dalign)
 		stripe_align = mp->m_dalign;
 
-	if (xfs_alloc_is_userdata(ap->datatype))
+	if (ap->flags & XFS_BMAPI_COWFORK)
+		align = xfs_get_cowextsz_hint(ap->ip);
+	else if (xfs_alloc_is_userdata(ap->datatype))
 		align = xfs_get_extsz_hint(ap->ip);
 	if (unlikely(align)) {
 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
@@ -3856,7 +3906,8 @@ xfs_bmap_btalloc(
 		ASSERT(nullfb || fb_agno == args.agno ||
 		       (ap->dfops->dop_low && fb_agno < args.agno));
 		ap->length = args.len;
-		ap->ip->i_d.di_nblocks += args.len;
+		if (!(ap->flags & XFS_BMAPI_COWFORK))
+			ap->ip->i_d.di_nblocks += args.len;
 		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
 		if (ap->wasdel)
 			ap->ip->i_delayed_blks -= args.len;
@@ -3875,6 +3926,63 @@ xfs_bmap_btalloc(
 	return 0;
 }
 
+/*
+ * For a remap operation, just "allocate" an extent at the address that the
+ * caller passed in, and ensure that the AGFL is the right size.  The caller
+ * will then map the "allocated" extent into the file somewhere.
+ */
+STATIC int
+xfs_bmap_remap_alloc(
+	struct xfs_bmalloca	*ap)
+{
+	struct xfs_trans	*tp = ap->tp;
+	struct xfs_mount	*mp = tp->t_mountp;
+	xfs_agblock_t		bno;
+	struct xfs_alloc_arg	args;
+	int			error;
+
+	/*
+	 * validate that the block number is legal - the enables us to detect
+	 * and handle a silent filesystem corruption rather than crashing.
+	 */
+	memset(&args, 0, sizeof(struct xfs_alloc_arg));
+	args.tp = ap->tp;
+	args.mp = ap->tp->t_mountp;
+	bno = *ap->firstblock;
+	args.agno = XFS_FSB_TO_AGNO(mp, bno);
+	args.agbno = XFS_FSB_TO_AGBNO(mp, bno);
+	if (args.agno >= mp->m_sb.sb_agcount ||
+	    args.agbno >= mp->m_sb.sb_agblocks)
+		return -EFSCORRUPTED;
+
+	/* "Allocate" the extent from the range we passed in. */
+	trace_xfs_bmap_remap_alloc(ap->ip, *ap->firstblock, ap->length);
+	ap->blkno = bno;
+	ap->ip->i_d.di_nblocks += ap->length;
+	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
+
+	/* Fix the freelist, like a real allocator does. */
+	args.datatype = ap->datatype;
+	args.pag = xfs_perag_get(args.mp, args.agno);
+	ASSERT(args.pag);
+
+	/*
+	 * The freelist fixing code will decline the allocation if
+	 * the size and shape of the free space doesn't allow for
+	 * allocating the extent and updating all the metadata that
+	 * happens during an allocation.  We're remapping, not
+	 * allocating, so skip that check by pretending to be freeing.
+	 */
+	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
+	if (error)
+		goto error0;
+error0:
+	xfs_perag_put(args.pag);
+	if (error)
+		trace_xfs_bmap_remap_alloc_error(ap->ip, error, _RET_IP_);
+	return error;
+}
+
 /*
  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  * It figures out where to ask the underlying allocator to put the new extent.
@@ -3883,6 +3991,8 @@ STATIC int
 xfs_bmap_alloc(
 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
 {
+	if (ap->flags & XFS_BMAPI_REMAP)
+		return xfs_bmap_remap_alloc(ap);
 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
 	    xfs_alloc_is_userdata(ap->datatype))
 		return xfs_bmap_rtalloc(ap);
@@ -4012,12 +4122,11 @@ xfs_bmapi_read(
 	int			error;
 	int			eof;
 	int			n = 0;
-	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
-						XFS_ATTR_FORK : XFS_DATA_FORK;
+	int			whichfork = xfs_bmapi_whichfork(flags);
 
 	ASSERT(*nmap >= 1);
 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
-			   XFS_BMAPI_IGSTATE)));
+			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
 
 	if (unlikely(XFS_TEST_ERROR(
@@ -4035,6 +4144,16 @@ xfs_bmapi_read(
 
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 
+	/* No CoW fork?  Return a hole. */
+	if (whichfork == XFS_COW_FORK && !ifp) {
+		mval->br_startoff = bno;
+		mval->br_startblock = HOLESTARTBLOCK;
+		mval->br_blockcount = len;
+		mval->br_state = XFS_EXT_NORM;
+		*nmap = 1;
+		return 0;
+	}
+
 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 		error = xfs_iread_extents(NULL, ip, whichfork);
 		if (error)
@@ -4084,6 +4203,7 @@ xfs_bmapi_read(
 int
 xfs_bmapi_reserve_delalloc(
 	struct xfs_inode	*ip,
+	int			whichfork,
 	xfs_fileoff_t		aoff,
 	xfs_filblks_t		len,
 	struct xfs_bmbt_irec	*got,
@@ -4092,7 +4212,7 @@ xfs_bmapi_reserve_delalloc(
 	int			eof)
 {
 	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
+	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
 	xfs_extlen_t		alen;
 	xfs_extlen_t		indlen;
 	char			rt = XFS_IS_REALTIME_INODE(ip);
@@ -4104,7 +4224,10 @@ xfs_bmapi_reserve_delalloc(
 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
 
 	/* Figure out the extent size, adjust alen */
-	extsz = xfs_get_extsz_hint(ip);
+	if (whichfork == XFS_COW_FORK)
+		extsz = xfs_get_cowextsz_hint(ip);
+	else
+		extsz = xfs_get_extsz_hint(ip);
 	if (extsz) {
 		error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
 					       1, 0, &aoff, &alen);
@@ -4151,7 +4274,7 @@ xfs_bmapi_reserve_delalloc(
 	got->br_startblock = nullstartblock(indlen);
 	got->br_blockcount = alen;
 	got->br_state = XFS_EXT_NORM;
-	xfs_bmap_add_extent_hole_delay(ip, lastx, got);
+	xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
 
 	/*
 	 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
@@ -4182,8 +4305,7 @@ xfs_bmapi_allocate(
 	struct xfs_bmalloca	*bma)
 {
 	struct xfs_mount	*mp = bma->ip->i_mount;
-	int			whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
-						XFS_ATTR_FORK : XFS_DATA_FORK;
+	int			whichfork = xfs_bmapi_whichfork(bma->flags);
 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
 	int			tmp_logflags = 0;
 	int			error;
@@ -4278,7 +4400,7 @@ xfs_bmapi_allocate(
 		bma->got.br_state = XFS_EXT_UNWRITTEN;
 
 	if (bma->wasdel)
-		error = xfs_bmap_add_extent_delay_real(bma);
+		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
 	else
 		error = xfs_bmap_add_extent_hole_real(bma, whichfork);
 
@@ -4308,8 +4430,7 @@ xfs_bmapi_convert_unwritten(
 	xfs_filblks_t		len,
 	int			flags)
 {
-	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
-						XFS_ATTR_FORK : XFS_DATA_FORK;
+	int			whichfork = xfs_bmapi_whichfork(flags);
 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
 	int			tmp_logflags = 0;
 	int			error;
@@ -4325,6 +4446,8 @@ xfs_bmapi_convert_unwritten(
 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
 		return 0;
 
+	ASSERT(whichfork != XFS_COW_FORK);
+
 	/*
 	 * Modify (by adding) the state flag, if writing.
 	 */
@@ -4431,8 +4554,7 @@ xfs_bmapi_write(
 	orig_mval = mval;
 	orig_nmap = *nmap;
 #endif
-	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
-		XFS_ATTR_FORK : XFS_DATA_FORK;
+	whichfork = xfs_bmapi_whichfork(flags);
 
 	ASSERT(*nmap >= 1);
 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
@@ -4441,6 +4563,11 @@ xfs_bmapi_write(
 	ASSERT(len > 0);
 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+	ASSERT(!(flags & XFS_BMAPI_REMAP) || whichfork == XFS_DATA_FORK);
+	ASSERT(!(flags & XFS_BMAPI_PREALLOC) || !(flags & XFS_BMAPI_REMAP));
+	ASSERT(!(flags & XFS_BMAPI_CONVERT) || !(flags & XFS_BMAPI_REMAP));
+	ASSERT(!(flags & XFS_BMAPI_PREALLOC) || whichfork != XFS_COW_FORK);
+	ASSERT(!(flags & XFS_BMAPI_CONVERT) || whichfork != XFS_COW_FORK);
 
 	/* zeroing is for currently only for data extents, not metadata */
 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
@@ -4501,6 +4628,14 @@ xfs_bmapi_write(
 		inhole = eof || bma.got.br_startoff > bno;
 		wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
 
+		/*
+		 * Make sure we only reflink into a hole.
+		 */
+		if (flags & XFS_BMAPI_REMAP)
+			ASSERT(inhole);
+		if (flags & XFS_BMAPI_COWFORK)
+			ASSERT(!inhole);
+
 		/*
 		 * First, deal with the hole before the allocated space
 		 * that we found, if any.
@@ -4531,6 +4666,17 @@ xfs_bmapi_write(
 				goto error0;
 			if (bma.blkno == NULLFSBLOCK)
 				break;
+
+			/*
+			 * If this is a CoW allocation, record the data in
+			 * the refcount btree for orphan recovery.
+			 */
+			if (whichfork == XFS_COW_FORK) {
+				error = xfs_refcount_alloc_cow_extent(mp, dfops,
+						bma.blkno, bma.length);
+				if (error)
+					goto error0;
+			}
 		}
 
 		/* Deal with the allocated space we found.  */
@@ -4696,7 +4842,8 @@ xfs_bmap_del_extent(
 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
 	int			*logflagsp, /* inode logging flags */
-	int			whichfork) /* data or attr fork */
+	int			whichfork, /* data or attr fork */
+	int			bflags)	/* bmapi flags */
 {
 	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
 	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
@@ -4725,6 +4872,8 @@ xfs_bmap_del_extent(
 
 	if (whichfork == XFS_ATTR_FORK)
 		state |= BMAP_ATTRFORK;
+	else if (whichfork == XFS_COW_FORK)
+		state |= BMAP_COWFORK;
 
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 	ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
@@ -4805,6 +4954,7 @@ xfs_bmap_del_extent(
 		/*
 		 * Matches the whole extent.  Delete the entry.
 		 */
+		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 		xfs_iext_remove(ip, *idx, 1,
 				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
 		--*idx;
@@ -4988,9 +5138,16 @@ xfs_bmap_del_extent(
 	/*
 	 * If we need to, add to list of extents to delete.
 	 */
-	if (do_fx)
-		xfs_bmap_add_free(mp, dfops, del->br_startblock,
-				del->br_blockcount, NULL);
+	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
+		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
+			error = xfs_refcount_decrease_extent(mp, dfops, del);
+			if (error)
+				goto done;
+		} else
+			xfs_bmap_add_free(mp, dfops, del->br_startblock,
+					del->br_blockcount, NULL);
+	}
+
 	/*
 	 * Adjust inode # blocks in the file.
 	 */
@@ -4999,7 +5156,7 @@ xfs_bmap_del_extent(
 	/*
 	 * Adjust quota data.
 	 */
-	if (qfield)
+	if (qfield && !(bflags & XFS_BMAPI_REMAP))
 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
 
 	/*
@@ -5014,6 +5171,175 @@ done:
 	return error;
 }
 
+/* Remove an extent from the CoW fork.  Similar to xfs_bmap_del_extent. */
+int
+xfs_bunmapi_cow(
+	struct xfs_inode		*ip,
+	struct xfs_bmbt_irec		*del)
+{
+	xfs_filblks_t			da_new;
+	xfs_filblks_t			da_old;
+	xfs_fsblock_t			del_endblock = 0;
+	xfs_fileoff_t			del_endoff;
+	int				delay;
+	struct xfs_bmbt_rec_host	*ep;
+	int				error;
+	struct xfs_bmbt_irec		got;
+	xfs_fileoff_t			got_endoff;
+	struct xfs_ifork		*ifp;
+	struct xfs_mount		*mp;
+	xfs_filblks_t			nblks;
+	struct xfs_bmbt_irec		new;
+	/* REFERENCED */
+	uint				qfield;
+	xfs_filblks_t			temp;
+	xfs_filblks_t			temp2;
+	int				state = BMAP_COWFORK;
+	int				eof;
+	xfs_extnum_t			eidx;
+
+	mp = ip->i_mount;
+	XFS_STATS_INC(mp, xs_del_exlist);
+
+	ep = xfs_bmap_search_extents(ip, del->br_startoff, XFS_COW_FORK, &eof,
+			&eidx, &got, &new);
+
+	ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); ifp = ifp;
+	ASSERT((eidx >= 0) && (eidx < ifp->if_bytes /
+		(uint)sizeof(xfs_bmbt_rec_t)));
+	ASSERT(del->br_blockcount > 0);
+	ASSERT(got.br_startoff <= del->br_startoff);
+	del_endoff = del->br_startoff + del->br_blockcount;
+	got_endoff = got.br_startoff + got.br_blockcount;
+	ASSERT(got_endoff >= del_endoff);
+	delay = isnullstartblock(got.br_startblock);
+	ASSERT(isnullstartblock(del->br_startblock) == delay);
+	qfield = 0;
+	error = 0;
+	/*
+	 * If deleting a real allocation, must free up the disk space.
+	 */
+	if (!delay) {
+		nblks = del->br_blockcount;
+		qfield = XFS_TRANS_DQ_BCOUNT;
+		/*
+		 * Set up del_endblock and cur for later.
+		 */
+		del_endblock = del->br_startblock + del->br_blockcount;
+		da_old = da_new = 0;
+	} else {
+		da_old = startblockval(got.br_startblock);
+		da_new = 0;
+		nblks = 0;
+	}
+	qfield = qfield;
+	nblks = nblks;
+
+	/*
+	 * Set flag value to use in switch statement.
+	 * Left-contig is 2, right-contig is 1.
+	 */
+	switch (((got.br_startoff == del->br_startoff) << 1) |
+		(got_endoff == del_endoff)) {
+	case 3:
+		/*
+		 * Matches the whole extent.  Delete the entry.
+		 */
+		xfs_iext_remove(ip, eidx, 1, BMAP_COWFORK);
+		--eidx;
+		break;
+
+	case 2:
+		/*
+		 * Deleting the first part of the extent.
+		 */
+		trace_xfs_bmap_pre_update(ip, eidx, state, _THIS_IP_);
+		xfs_bmbt_set_startoff(ep, del_endoff);
+		temp = got.br_blockcount - del->br_blockcount;
+		xfs_bmbt_set_blockcount(ep, temp);
+		if (delay) {
+			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
+				da_old);
+			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
+			trace_xfs_bmap_post_update(ip, eidx, state, _THIS_IP_);
+			da_new = temp;
+			break;
+		}
+		xfs_bmbt_set_startblock(ep, del_endblock);
+		trace_xfs_bmap_post_update(ip, eidx, state, _THIS_IP_);
+		break;
+
+	case 1:
+		/*
+		 * Deleting the last part of the extent.
+		 */
+		temp = got.br_blockcount - del->br_blockcount;
+		trace_xfs_bmap_pre_update(ip, eidx, state, _THIS_IP_);
+		xfs_bmbt_set_blockcount(ep, temp);
+		if (delay) {
+			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
+				da_old);
+			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
+			trace_xfs_bmap_post_update(ip, eidx, state, _THIS_IP_);
+			da_new = temp;
+			break;
+		}
+		trace_xfs_bmap_post_update(ip, eidx, state, _THIS_IP_);
+		break;
+
+	case 0:
+		/*
+		 * Deleting the middle of the extent.
+		 */
+		temp = del->br_startoff - got.br_startoff;
+		trace_xfs_bmap_pre_update(ip, eidx, state, _THIS_IP_);
+		xfs_bmbt_set_blockcount(ep, temp);
+		new.br_startoff = del_endoff;
+		temp2 = got_endoff - del_endoff;
+		new.br_blockcount = temp2;
+		new.br_state = got.br_state;
+		if (!delay) {
+			new.br_startblock = del_endblock;
+		} else {
+			temp = xfs_bmap_worst_indlen(ip, temp);
+			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
+			temp2 = xfs_bmap_worst_indlen(ip, temp2);
+			new.br_startblock = nullstartblock((int)temp2);
+			da_new = temp + temp2;
+			while (da_new > da_old) {
+				if (temp) {
+					temp--;
+					da_new--;
+					xfs_bmbt_set_startblock(ep,
+						nullstartblock((int)temp));
+				}
+				if (da_new == da_old)
+					break;
+				if (temp2) {
+					temp2--;
+					da_new--;
+					new.br_startblock =
+						nullstartblock((int)temp2);
+				}
+			}
+		}
+		trace_xfs_bmap_post_update(ip, eidx, state, _THIS_IP_);
+		xfs_iext_insert(ip, eidx + 1, 1, &new, state);
+		++eidx;
+		break;
+	}
+
+	/*
+	 * Account for change in delayed indirect blocks.
+	 * Nothing to do for disk quota accounting here.
+	 */
+	ASSERT(da_old >= da_new);
+	if (da_old > da_new)
+		xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
+
+	return error;
+}
+
 /*
  * Unmap (remove) blocks from a file.
  * If nexts is nonzero then the number of extents to remove is limited to
@@ -5021,17 +5347,16 @@ done:
  * *done is set.
  */
 int						/* error */
-xfs_bunmapi(
+__xfs_bunmapi(
 	xfs_trans_t		*tp,		/* transaction pointer */
 	struct xfs_inode	*ip,		/* incore inode */
 	xfs_fileoff_t		bno,		/* starting offset to unmap */
-	xfs_filblks_t		len,		/* length to unmap in file */
+	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
 	int			flags,		/* misc flags */
 	xfs_extnum_t		nexts,		/* number of extents max */
 	xfs_fsblock_t		*firstblock,	/* first allocated block
 						   controls a.g. for allocs */
-	struct xfs_defer_ops	*dfops,		/* i/o: list extents to free */
-	int			*done)		/* set if not done yet */
+	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
 {
 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
 	xfs_bmbt_irec_t		del;		/* extent being deleted */
@@ -5053,11 +5378,12 @@ xfs_bunmapi(
 	int			wasdel;		/* was a delayed alloc extent */
 	int			whichfork;	/* data or attribute fork */
 	xfs_fsblock_t		sum;
+	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
 
 	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
 
-	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
-		XFS_ATTR_FORK : XFS_DATA_FORK;
+	whichfork = xfs_bmapi_whichfork(flags);
+	ASSERT(whichfork != XFS_COW_FORK);
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 	if (unlikely(
 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
@@ -5079,7 +5405,7 @@ xfs_bunmapi(
 		return error;
 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 	if (nextents == 0) {
-		*done = 1;
+		*rlen = 0;
 		return 0;
 	}
 	XFS_STATS_INC(mp, xs_blk_unmap);
@@ -5324,7 +5650,7 @@ xfs_bunmapi(
 			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
 
 		error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
-				&tmp_logflags, whichfork);
+				&tmp_logflags, whichfork, flags);
 		logflags |= tmp_logflags;
 		if (error)
 			goto error0;
@@ -5350,7 +5676,10 @@ nodelete:
 			extno++;
 		}
 	}
-	*done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
+	if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
+		*rlen = 0;
+	else
+		*rlen = bno - start + 1;
 
 	/*
 	 * Convert to a btree if necessary.
@@ -5406,6 +5735,27 @@ error0:
 	return error;
 }
 
+/* Unmap a range of a file. */
+int
+xfs_bunmapi(
+	xfs_trans_t		*tp,
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		bno,
+	xfs_filblks_t		len,
+	int			flags,
+	xfs_extnum_t		nexts,
+	xfs_fsblock_t		*firstblock,
+	struct xfs_defer_ops	*dfops,
+	int			*done)
+{
+	int			error;
+
+	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
+			dfops);
+	*done = (len == 0);
+	return error;
+}
+
 /*
  * Determine whether an extent shift can be accomplished by a merge with the
  * extent that precedes the target hole of the shift.
@@ -5985,3 +6335,146 @@ out:
 	xfs_trans_cancel(tp);
 	return error;
 }
+
+/* Deferred mapping is only for real extents in the data fork. */
+static bool
+xfs_bmap_is_update_needed(
+	struct xfs_bmbt_irec	*bmap)
+{
+	return  bmap->br_startblock != HOLESTARTBLOCK &&
+		bmap->br_startblock != DELAYSTARTBLOCK;
+}
+
+/* Record a bmap intent. */
+static int
+__xfs_bmap_add(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	enum xfs_bmap_intent_type	type,
+	struct xfs_inode		*ip,
+	int				whichfork,
+	struct xfs_bmbt_irec		*bmap)
+{
+	int				error;
+	struct xfs_bmap_intent		*bi;
+
+	trace_xfs_bmap_defer(mp,
+			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
+			type,
+			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
+			ip->i_ino, whichfork,
+			bmap->br_startoff,
+			bmap->br_blockcount,
+			bmap->br_state);
+
+	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
+	INIT_LIST_HEAD(&bi->bi_list);
+	bi->bi_type = type;
+	bi->bi_owner = ip;
+	bi->bi_whichfork = whichfork;
+	bi->bi_bmap = *bmap;
+
+	error = xfs_defer_join(dfops, bi->bi_owner);
+	if (error) {
+		kmem_free(bi);
+		return error;
+	}
+
+	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
+	return 0;
+}
+
+/* Map an extent into a file. */
+int
+xfs_bmap_map_extent(
+	struct xfs_mount	*mp,
+	struct xfs_defer_ops	*dfops,
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*PREV)
+{
+	if (!xfs_bmap_is_update_needed(PREV))
+		return 0;
+
+	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
+			XFS_DATA_FORK, PREV);
+}
+
+/* Unmap an extent out of a file. */
+int
+xfs_bmap_unmap_extent(
+	struct xfs_mount	*mp,
+	struct xfs_defer_ops	*dfops,
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*PREV)
+{
+	if (!xfs_bmap_is_update_needed(PREV))
+		return 0;
+
+	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
+			XFS_DATA_FORK, PREV);
+}
+
+/*
+ * Process one of the deferred bmap operations.  We pass back the
+ * btree cursor to maintain our lock on the bmapbt between calls.
+ */
+int
+xfs_bmap_finish_one(
+	struct xfs_trans		*tp,
+	struct xfs_defer_ops		*dfops,
+	struct xfs_inode		*ip,
+	enum xfs_bmap_intent_type	type,
+	int				whichfork,
+	xfs_fileoff_t			startoff,
+	xfs_fsblock_t			startblock,
+	xfs_filblks_t			blockcount,
+	xfs_exntst_t			state)
+{
+	struct xfs_bmbt_irec		bmap;
+	int				nimaps = 1;
+	xfs_fsblock_t			firstfsb;
+	int				flags = XFS_BMAPI_REMAP;
+	int				done;
+	int				error = 0;
+
+	bmap.br_startblock = startblock;
+	bmap.br_startoff = startoff;
+	bmap.br_blockcount = blockcount;
+	bmap.br_state = state;
+
+	trace_xfs_bmap_deferred(tp->t_mountp,
+			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
+			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
+			ip->i_ino, whichfork, startoff, blockcount, state);
+
+	if (whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK)
+		return -EFSCORRUPTED;
+	if (whichfork == XFS_ATTR_FORK)
+		flags |= XFS_BMAPI_ATTRFORK;
+
+	if (XFS_TEST_ERROR(false, tp->t_mountp,
+			XFS_ERRTAG_BMAP_FINISH_ONE,
+			XFS_RANDOM_BMAP_FINISH_ONE))
+		return -EIO;
+
+	switch (type) {
+	case XFS_BMAP_MAP:
+		firstfsb = bmap.br_startblock;
+		error = xfs_bmapi_write(tp, ip, bmap.br_startoff,
+					bmap.br_blockcount, flags, &firstfsb,
+					bmap.br_blockcount, &bmap, &nimaps,
+					dfops);
+		break;
+	case XFS_BMAP_UNMAP:
+		error = xfs_bunmapi(tp, ip, bmap.br_startoff,
+				bmap.br_blockcount, flags, 1, &firstfsb,
+				dfops, &done);
+		ASSERT(done);
+		break;
+	default:
+		ASSERT(0);
+		error = -EFSCORRUPTED;
+	}
+
+	return error;
+}

fs/xfs/libxfs/xfs_bmap.h

@@ -97,6 +97,19 @@ struct xfs_extent_free_item
  */
 #define XFS_BMAPI_ZERO		0x080
 
+/*
+ * Map the inode offset to the block given in ap->firstblock.  Primarily
+ * used for reflink.  The range must be in a hole, and this flag cannot be
+ * turned on with PREALLOC or CONVERT, and cannot be used on the attr fork.
+ *
+ * For bunmapi, this flag unmaps the range without adjusting quota, reducing
+ * refcount, or freeing the blocks.
+ */
+#define XFS_BMAPI_REMAP		0x100
+
+/* Map something in the CoW fork. */
+#define XFS_BMAPI_COWFORK	0x200
+
 #define XFS_BMAPI_FLAGS \
 	{ XFS_BMAPI_ENTIRE,	"ENTIRE" }, \
 	{ XFS_BMAPI_METADATA,	"METADATA" }, \
@@ -105,12 +118,24 @@ struct xfs_extent_free_item
 	{ XFS_BMAPI_IGSTATE,	"IGSTATE" }, \
 	{ XFS_BMAPI_CONTIG,	"CONTIG" }, \
 	{ XFS_BMAPI_CONVERT,	"CONVERT" }, \
-	{ XFS_BMAPI_ZERO,	"ZERO" }
+	{ XFS_BMAPI_ZERO,	"ZERO" }, \
+	{ XFS_BMAPI_REMAP,	"REMAP" }, \
+	{ XFS_BMAPI_COWFORK,	"COWFORK" }
 
 
 static inline int xfs_bmapi_aflag(int w)
 {
-	return (w == XFS_ATTR_FORK ? XFS_BMAPI_ATTRFORK : 0);
+	return (w == XFS_ATTR_FORK ? XFS_BMAPI_ATTRFORK :
+	       (w == XFS_COW_FORK ? XFS_BMAPI_COWFORK : 0));
+}
+
+static inline int xfs_bmapi_whichfork(int bmapi_flags)
+{
+	if (bmapi_flags & XFS_BMAPI_COWFORK)
+		return XFS_COW_FORK;
+	else if (bmapi_flags & XFS_BMAPI_ATTRFORK)
+		return XFS_ATTR_FORK;
+	return XFS_DATA_FORK;
 }
 
 /*
@@ -131,13 +156,15 @@ static inline int xfs_bmapi_aflag(int w)
 #define BMAP_LEFT_VALID		(1 << 6)
 #define BMAP_RIGHT_VALID	(1 << 7)
 #define BMAP_ATTRFORK		(1 << 8)
+#define BMAP_COWFORK		(1 << 9)
 
 #define XFS_BMAP_EXT_FLAGS \
 	{ BMAP_LEFT_CONTIG,	"LC" }, \
 	{ BMAP_RIGHT_CONTIG,	"RC" }, \
 	{ BMAP_LEFT_FILLING,	"LF" }, \
 	{ BMAP_RIGHT_FILLING,	"RF" }, \
-	{ BMAP_ATTRFORK,	"ATTR" }
+	{ BMAP_ATTRFORK,	"ATTR" }, \
+	{ BMAP_COWFORK,		"COW" }
 
 
 /*
@@ -186,10 +213,15 @@ int	xfs_bmapi_write(struct xfs_trans *tp, struct xfs_inode *ip,
 		xfs_fsblock_t *firstblock, xfs_extlen_t total,
 		struct xfs_bmbt_irec *mval, int *nmap,
 		struct xfs_defer_ops *dfops);
+int	__xfs_bunmapi(struct xfs_trans *tp, struct xfs_inode *ip,
+		xfs_fileoff_t bno, xfs_filblks_t *rlen, int flags,
+		xfs_extnum_t nexts, xfs_fsblock_t *firstblock,
+		struct xfs_defer_ops *dfops);
 int	xfs_bunmapi(struct xfs_trans *tp, struct xfs_inode *ip,
 		xfs_fileoff_t bno, xfs_filblks_t len, int flags,
 		xfs_extnum_t nexts, xfs_fsblock_t *firstblock,
 		struct xfs_defer_ops *dfops, int *done);
+int	xfs_bunmapi_cow(struct xfs_inode *ip, struct xfs_bmbt_irec *del);
 int	xfs_check_nostate_extents(struct xfs_ifork *ifp, xfs_extnum_t idx,
 		xfs_extnum_t num);
 uint	xfs_default_attroffset(struct xfs_inode *ip);
@@ -203,8 +235,31 @@ struct xfs_bmbt_rec_host *
 	xfs_bmap_search_extents(struct xfs_inode *ip, xfs_fileoff_t bno,
 		int fork, int *eofp, xfs_extnum_t *lastxp,
 		struct xfs_bmbt_irec *gotp, struct xfs_bmbt_irec *prevp);
-int	xfs_bmapi_reserve_delalloc(struct xfs_inode *ip, xfs_fileoff_t aoff,
-		xfs_filblks_t len, struct xfs_bmbt_irec *got,
-		struct xfs_bmbt_irec *prev, xfs_extnum_t *lastx, int eof);
+int	xfs_bmapi_reserve_delalloc(struct xfs_inode *ip, int whichfork,
+		xfs_fileoff_t aoff, xfs_filblks_t len,
+		struct xfs_bmbt_irec *got, struct xfs_bmbt_irec *prev,
+		xfs_extnum_t *lastx, int eof);
+
+enum xfs_bmap_intent_type {
+	XFS_BMAP_MAP = 1,
+	XFS_BMAP_UNMAP,
+};
+
+struct xfs_bmap_intent {
+	struct list_head			bi_list;
+	enum xfs_bmap_intent_type		bi_type;
+	struct xfs_inode			*bi_owner;
+	int					bi_whichfork;
+	struct xfs_bmbt_irec			bi_bmap;
+};
+
+int	xfs_bmap_finish_one(struct xfs_trans *tp, struct xfs_defer_ops *dfops,
+		struct xfs_inode *ip, enum xfs_bmap_intent_type type,
+		int whichfork, xfs_fileoff_t startoff, xfs_fsblock_t startblock,
+		xfs_filblks_t blockcount, xfs_exntst_t state);
+int	xfs_bmap_map_extent(struct xfs_mount *mp, struct xfs_defer_ops *dfops,
+		struct xfs_inode *ip, struct xfs_bmbt_irec *imap);
+int	xfs_bmap_unmap_extent(struct xfs_mount *mp, struct xfs_defer_ops *dfops,
+		struct xfs_inode *ip, struct xfs_bmbt_irec *imap);
 
 #endif	/* __XFS_BMAP_H__ */

fs/xfs/libxfs/xfs_bmap_btree.c

@@ -453,6 +453,7 @@ xfs_bmbt_alloc_block(
 
 	if (args.fsbno == NULLFSBLOCK) {
 		args.fsbno = be64_to_cpu(start->l);
+try_another_ag:
 		args.type = XFS_ALLOCTYPE_START_BNO;
 		/*
 		 * Make sure there is sufficient room left in the AG to
@@ -482,6 +483,22 @@ xfs_bmbt_alloc_block(
 	if (error)
 		goto error0;
 
+	/*
+	 * During a CoW operation, the allocation and bmbt updates occur in
+	 * different transactions.  The mapping code tries to put new bmbt
+	 * blocks near extents being mapped, but the only way to guarantee this
+	 * is if the alloc and the mapping happen in a single transaction that
+	 * has a block reservation.  That isn't the case here, so if we run out
+	 * of space we'll try again with another AG.
+	 */
+	if (xfs_sb_version_hasreflink(&cur->bc_mp->m_sb) &&
+	    args.fsbno == NULLFSBLOCK &&
+	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
+		cur->bc_private.b.dfops->dop_low = true;
+		args.fsbno = cur->bc_private.b.firstblock;
+		goto try_another_ag;
+	}
+
 	if (args.fsbno == NULLFSBLOCK && args.minleft) {
 		/*
 		 * Could not find an AG with enough free space to satisfy
@@ -777,6 +794,7 @@ xfs_bmbt_init_cursor(
 {
 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
 	struct xfs_btree_cur	*cur;
+	ASSERT(whichfork != XFS_COW_FORK);
 
 	cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_SLEEP);
 

fs/xfs/libxfs/xfs_btree.c

@@ -45,9 +45,10 @@ kmem_zone_t	*xfs_btree_cur_zone;
  */
 static const __uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
 	{ XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
-	  XFS_FIBT_MAGIC },
+	  XFS_FIBT_MAGIC, 0 },
 	{ XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
-	  XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC }
+	  XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
+	  XFS_REFC_CRC_MAGIC }
 };
 #define xfs_btree_magic(cur) \
 	xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
@@ -1216,6 +1217,9 @@ xfs_btree_set_refs(
 	case XFS_BTNUM_RMAP:
 		xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
 		break;
+	case XFS_BTNUM_REFC:
+		xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
+		break;
 	default:
 		ASSERT(0);
 	}

fs/xfs/libxfs/xfs_btree.h

@@ -49,6 +49,7 @@ union xfs_btree_key {
 	struct xfs_inobt_key		inobt;
 	struct xfs_rmap_key		rmap;
 	struct xfs_rmap_key		__rmap_bigkey[2];
+	struct xfs_refcount_key		refc;
 };
 
 union xfs_btree_rec {
@@ -57,6 +58,7 @@ union xfs_btree_rec {
 	struct xfs_alloc_rec		alloc;
 	struct xfs_inobt_rec		inobt;
 	struct xfs_rmap_rec		rmap;
+	struct xfs_refcount_rec		refc;
 };
 
 /*
@@ -72,6 +74,7 @@ union xfs_btree_rec {
 #define	XFS_BTNUM_INO	((xfs_btnum_t)XFS_BTNUM_INOi)
 #define	XFS_BTNUM_FINO	((xfs_btnum_t)XFS_BTNUM_FINOi)
 #define	XFS_BTNUM_RMAP	((xfs_btnum_t)XFS_BTNUM_RMAPi)
+#define	XFS_BTNUM_REFC	((xfs_btnum_t)XFS_BTNUM_REFCi)
 
 /*
  * For logging record fields.
@@ -105,6 +108,7 @@ do {    \
 	case XFS_BTNUM_INO: __XFS_BTREE_STATS_INC(__mp, ibt, stat); break; \
 	case XFS_BTNUM_FINO: __XFS_BTREE_STATS_INC(__mp, fibt, stat); break; \
 	case XFS_BTNUM_RMAP: __XFS_BTREE_STATS_INC(__mp, rmap, stat); break; \
+	case XFS_BTNUM_REFC: __XFS_BTREE_STATS_INC(__mp, refcbt, stat); break; \
 	case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break;	\
 	}       \
 } while (0)
@@ -127,6 +131,8 @@ do {    \
 		__XFS_BTREE_STATS_ADD(__mp, fibt, stat, val); break; \
 	case XFS_BTNUM_RMAP:	\
 		__XFS_BTREE_STATS_ADD(__mp, rmap, stat, val); break; \
+	case XFS_BTNUM_REFC:	\
+		__XFS_BTREE_STATS_ADD(__mp, refcbt, stat, val); break; \
 	case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break; \
 	}       \
 } while (0)
@@ -217,6 +223,15 @@ union xfs_btree_irec {
 	struct xfs_bmbt_irec		b;
 	struct xfs_inobt_rec_incore	i;
 	struct xfs_rmap_irec		r;
+	struct xfs_refcount_irec	rc;
+};
+
+/* Per-AG btree private information. */
+union xfs_btree_cur_private {
+	struct {
+		unsigned long	nr_ops;		/* # record updates */
+		int		shape_changes;	/* # of extent splits */
+	} refc;
 };
 
 /*
@@ -243,6 +258,7 @@ typedef struct xfs_btree_cur
 			struct xfs_buf	*agbp;	/* agf/agi buffer pointer */
 			struct xfs_defer_ops *dfops;	/* deferred updates */
 			xfs_agnumber_t	agno;	/* ag number */
+			union xfs_btree_cur_private	priv;
 		} a;
 		struct {			/* needed for BMAP */
 			struct xfs_inode *ip;	/* pointer to our inode */

fs/xfs/libxfs/xfs_defer.h

@@ -51,6 +51,8 @@ struct xfs_defer_pending {
  * find all the space it needs.
  */
 enum xfs_defer_ops_type {
+	XFS_DEFER_OPS_TYPE_BMAP,
+	XFS_DEFER_OPS_TYPE_REFCOUNT,
 	XFS_DEFER_OPS_TYPE_RMAP,
 	XFS_DEFER_OPS_TYPE_FREE,
 	XFS_DEFER_OPS_TYPE_MAX,

fs/xfs/libxfs/xfs_format.h

@@ -456,9 +456,11 @@ xfs_sb_has_compat_feature(
 
 #define XFS_SB_FEAT_RO_COMPAT_FINOBT   (1 << 0)		/* free inode btree */
 #define XFS_SB_FEAT_RO_COMPAT_RMAPBT   (1 << 1)		/* reverse map btree */
+#define XFS_SB_FEAT_RO_COMPAT_REFLINK  (1 << 2)		/* reflinked files */
 #define XFS_SB_FEAT_RO_COMPAT_ALL \
 		(XFS_SB_FEAT_RO_COMPAT_FINOBT | \
-		 XFS_SB_FEAT_RO_COMPAT_RMAPBT)
+		 XFS_SB_FEAT_RO_COMPAT_RMAPBT | \
+		 XFS_SB_FEAT_RO_COMPAT_REFLINK)
 #define XFS_SB_FEAT_RO_COMPAT_UNKNOWN	~XFS_SB_FEAT_RO_COMPAT_ALL
 static inline bool
 xfs_sb_has_ro_compat_feature(
@@ -546,6 +548,12 @@ static inline bool xfs_sb_version_hasrmapbt(struct xfs_sb *sbp)
 		(sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT);
 }
 
+static inline bool xfs_sb_version_hasreflink(struct xfs_sb *sbp)
+{
+	return XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
+		(sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK);
+}
+
 /*
  * end of superblock version macros
  */
@@ -641,14 +649,17 @@ typedef struct xfs_agf {
 	uuid_t		agf_uuid;	/* uuid of filesystem */
 
 	__be32		agf_rmap_blocks;	/* rmapbt blocks used */
-	__be32		agf_padding;		/* padding */
+	__be32		agf_refcount_blocks;	/* refcountbt blocks used */
+
+	__be32		agf_refcount_root;	/* refcount tree root block */
+	__be32		agf_refcount_level;	/* refcount btree levels */
 
 	/*
 	 * reserve some contiguous space for future logged fields before we add
 	 * the unlogged fields. This makes the range logging via flags and
 	 * structure offsets much simpler.
 	 */
-	__be64		agf_spare64[15];
+	__be64		agf_spare64[14];
 
 	/* unlogged fields, written during buffer writeback. */
 	__be64		agf_lsn;	/* last write sequence */
@@ -674,8 +685,11 @@ typedef struct xfs_agf {
 #define	XFS_AGF_BTREEBLKS	0x00000800
 #define	XFS_AGF_UUID		0x00001000
 #define	XFS_AGF_RMAP_BLOCKS	0x00002000
-#define	XFS_AGF_SPARE64		0x00004000
-#define	XFS_AGF_NUM_BITS	15
+#define	XFS_AGF_REFCOUNT_BLOCKS	0x00004000
+#define	XFS_AGF_REFCOUNT_ROOT	0x00008000
+#define	XFS_AGF_REFCOUNT_LEVEL	0x00010000
+#define	XFS_AGF_SPARE64		0x00020000
+#define	XFS_AGF_NUM_BITS	18
 #define	XFS_AGF_ALL_BITS	((1 << XFS_AGF_NUM_BITS) - 1)
 
 #define XFS_AGF_FLAGS \
@@ -693,6 +707,9 @@ typedef struct xfs_agf {
 	{ XFS_AGF_BTREEBLKS,	"BTREEBLKS" }, \
 	{ XFS_AGF_UUID,		"UUID" }, \
 	{ XFS_AGF_RMAP_BLOCKS,	"RMAP_BLOCKS" }, \
+	{ XFS_AGF_REFCOUNT_BLOCKS,	"REFCOUNT_BLOCKS" }, \
+	{ XFS_AGF_REFCOUNT_ROOT,	"REFCOUNT_ROOT" }, \
+	{ XFS_AGF_REFCOUNT_LEVEL,	"REFCOUNT_LEVEL" }, \
 	{ XFS_AGF_SPARE64,	"SPARE64" }
 
 /* disk block (xfs_daddr_t) in the AG */
@@ -885,7 +902,8 @@ typedef struct xfs_dinode {
 	__be64		di_changecount;	/* number of attribute changes */
 	__be64		di_lsn;		/* flush sequence */
 	__be64		di_flags2;	/* more random flags */
-	__u8		di_pad2[16];	/* more padding for future expansion */
+	__be32		di_cowextsize;	/* basic cow extent size for file */
+	__u8		di_pad2[12];	/* more padding for future expansion */
 
 	/* fields only written to during inode creation */
 	xfs_timestamp_t	di_crtime;	/* time created */
@@ -1041,9 +1059,14 @@ static inline void xfs_dinode_put_rdev(struct xfs_dinode *dip, xfs_dev_t rdev)
  * 16 bits of the XFS_XFLAG_s range.
  */
 #define XFS_DIFLAG2_DAX_BIT	0	/* use DAX for this inode */
+#define XFS_DIFLAG2_REFLINK_BIT	1	/* file's blocks may be shared */
+#define XFS_DIFLAG2_COWEXTSIZE_BIT   2  /* copy on write extent size hint */
 #define XFS_DIFLAG2_DAX		(1 << XFS_DIFLAG2_DAX_BIT)
+#define XFS_DIFLAG2_REFLINK     (1 << XFS_DIFLAG2_REFLINK_BIT)
+#define XFS_DIFLAG2_COWEXTSIZE  (1 << XFS_DIFLAG2_COWEXTSIZE_BIT)
 
-#define XFS_DIFLAG2_ANY		(XFS_DIFLAG2_DAX)
+#define XFS_DIFLAG2_ANY \
+	(XFS_DIFLAG2_DAX | XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)
 
 /*
  * Inode number format:
@@ -1353,7 +1376,9 @@ struct xfs_owner_info {
 #define XFS_RMAP_OWN_AG		(-5ULL)	/* AG freespace btree blocks */
 #define XFS_RMAP_OWN_INOBT	(-6ULL)	/* Inode btree blocks */
 #define XFS_RMAP_OWN_INODES	(-7ULL)	/* Inode chunk */
-#define XFS_RMAP_OWN_MIN	(-8ULL) /* guard */
+#define XFS_RMAP_OWN_REFC	(-8ULL) /* refcount tree */
+#define XFS_RMAP_OWN_COW	(-9ULL) /* cow allocations */
+#define XFS_RMAP_OWN_MIN	(-10ULL) /* guard */
 
 #define XFS_RMAP_NON_INODE_OWNER(owner)	(!!((owner) & (1ULL << 63)))
 
@@ -1433,6 +1458,62 @@ typedef __be32 xfs_rmap_ptr_t;
 	 XFS_FIBT_BLOCK(mp) + 1 : \
 	 XFS_IBT_BLOCK(mp) + 1)
 
+/*
+ * Reference Count Btree format definitions
+ *
+ */
+#define	XFS_REFC_CRC_MAGIC	0x52334643	/* 'R3FC' */
+
+unsigned int xfs_refc_block(struct xfs_mount *mp);
+
+/*
+ * Data record/key structure
+ *
+ * Each record associates a range of physical blocks (starting at
+ * rc_startblock and ending rc_blockcount blocks later) with a reference
+ * count (rc_refcount).  Extents that are being used to stage a copy on
+ * write (CoW) operation are recorded in the refcount btree with a
+ * refcount of 1.  All other records must have a refcount > 1 and must
+ * track an extent mapped only by file data forks.
+ *
+ * Extents with a single owner (attributes, metadata, non-shared file
+ * data) are not tracked here.  Free space is also not tracked here.
+ * This is consistent with pre-reflink XFS.
+ */
+
+/*
+ * Extents that are being used to stage a copy on write are stored
+ * in the refcount btree with a refcount of 1 and the upper bit set
+ * on the startblock.  This speeds up mount time deletion of stale
+ * staging extents because they're all at the right side of the tree.
+ */
+#define XFS_REFC_COW_START		((xfs_agblock_t)(1U << 31))
+#define REFCNTBT_COWFLAG_BITLEN		1
+#define REFCNTBT_AGBLOCK_BITLEN		31
+
+struct xfs_refcount_rec {
+	__be32		rc_startblock;	/* starting block number */
+	__be32		rc_blockcount;	/* count of blocks */
+	__be32		rc_refcount;	/* number of inodes linked here */
+};
+
+struct xfs_refcount_key {
+	__be32		rc_startblock;	/* starting block number */
+};
+
+struct xfs_refcount_irec {
+	xfs_agblock_t	rc_startblock;	/* starting block number */
+	xfs_extlen_t	rc_blockcount;	/* count of free blocks */
+	xfs_nlink_t	rc_refcount;	/* number of inodes linked here */
+};
+
+#define MAXREFCOUNT	((xfs_nlink_t)~0U)
+#define MAXREFCEXTLEN	((xfs_extlen_t)~0U)
+
+/* btree pointer type */
+typedef __be32 xfs_refcount_ptr_t;
+
+
 /*
  * BMAP Btree format definitions
  *

fs/xfs/libxfs/xfs_fs.h

@@ -81,14 +81,16 @@ struct getbmapx {
 #define BMV_IF_PREALLOC		0x4	/* rtn status BMV_OF_PREALLOC if req */
 #define BMV_IF_DELALLOC		0x8	/* rtn status BMV_OF_DELALLOC if req */
 #define BMV_IF_NO_HOLES		0x10	/* Do not return holes */
+#define BMV_IF_COWFORK		0x20	/* return CoW fork rather than data */
 #define BMV_IF_VALID	\
 	(BMV_IF_ATTRFORK|BMV_IF_NO_DMAPI_READ|BMV_IF_PREALLOC|	\
-	 BMV_IF_DELALLOC|BMV_IF_NO_HOLES)
+	 BMV_IF_DELALLOC|BMV_IF_NO_HOLES|BMV_IF_COWFORK)
 
 /*	bmv_oflags values - returned for each non-header segment */
 #define BMV_OF_PREALLOC		0x1	/* segment = unwritten pre-allocation */
 #define BMV_OF_DELALLOC		0x2	/* segment = delayed allocation */
 #define BMV_OF_LAST		0x4	/* segment is the last in the file */
+#define BMV_OF_SHARED		0x8	/* segment shared with another file */
 
 /*
  * Structure for XFS_IOC_FSSETDM.
@@ -206,7 +208,8 @@ typedef struct xfs_fsop_resblks {
 #define XFS_FSOP_GEOM_FLAGS_FTYPE	0x10000	/* inode directory types */
 #define XFS_FSOP_GEOM_FLAGS_FINOBT	0x20000	/* free inode btree */
 #define XFS_FSOP_GEOM_FLAGS_SPINODES	0x40000	/* sparse inode chunks	*/
-#define XFS_FSOP_GEOM_FLAGS_RMAPBT	0x80000	/* Reverse mapping btree */
+#define XFS_FSOP_GEOM_FLAGS_RMAPBT	0x80000	/* reverse mapping btree */
+#define XFS_FSOP_GEOM_FLAGS_REFLINK	0x100000 /* files can share blocks */
 
 /*
  * Minimum and maximum sizes need for growth checks.
@@ -275,7 +278,8 @@ typedef struct xfs_bstat {
 #define	bs_projid	bs_projid_lo	/* (previously just bs_projid)	*/
 	__u16		bs_forkoff;	/* inode fork offset in bytes	*/
 	__u16		bs_projid_hi;	/* higher part of project id	*/
-	unsigned char	bs_pad[10];	/* pad space, unused		*/
+	unsigned char	bs_pad[6];	/* pad space, unused		*/
+	__u32		bs_cowextsize;	/* cow extent size		*/
 	__u32		bs_dmevmask;	/* DMIG event mask		*/
 	__u16		bs_dmstate;	/* DMIG state info		*/
 	__u16		bs_aextents;	/* attribute number of extents	*/

fs/xfs/libxfs/xfs_inode_buf.c

@@ -256,6 +256,7 @@ xfs_inode_from_disk(
 		to->di_crtime.t_sec = be32_to_cpu(from->di_crtime.t_sec);
 		to->di_crtime.t_nsec = be32_to_cpu(from->di_crtime.t_nsec);
 		to->di_flags2 = be64_to_cpu(from->di_flags2);
+		to->di_cowextsize = be32_to_cpu(from->di_cowextsize);
 	}
 }
 
@@ -305,7 +306,7 @@ xfs_inode_to_disk(
 		to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
 		to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
 		to->di_flags2 = cpu_to_be64(from->di_flags2);
-
+		to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
 		to->di_ino = cpu_to_be64(ip->i_ino);
 		to->di_lsn = cpu_to_be64(lsn);
 		memset(to->di_pad2, 0, sizeof(to->di_pad2));
@@ -357,6 +358,7 @@ xfs_log_dinode_to_disk(
 		to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
 		to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
 		to->di_flags2 = cpu_to_be64(from->di_flags2);
+		to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
 		to->di_ino = cpu_to_be64(from->di_ino);
 		to->di_lsn = cpu_to_be64(from->di_lsn);
 		memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
@@ -373,6 +375,9 @@ xfs_dinode_verify(
 	struct xfs_inode	*ip,
 	struct xfs_dinode	*dip)
 {
+	uint16_t		flags;
+	uint64_t		flags2;
+
 	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
 		return false;
 
@@ -389,6 +394,23 @@ xfs_dinode_verify(
 		return false;
 	if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
 		return false;
+
+	flags = be16_to_cpu(dip->di_flags);
+	flags2 = be64_to_cpu(dip->di_flags2);
+
+	/* don't allow reflink/cowextsize if we don't have reflink */
+	if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
+            !xfs_sb_version_hasreflink(&mp->m_sb))
+		return false;
+
+	/* don't let reflink and realtime mix */
+	if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
+		return false;
+
+	/* don't let reflink and dax mix */
+	if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags2 & XFS_DIFLAG2_DAX))
+		return false;
+
 	return true;
 }
 

fs/xfs/libxfs/xfs_inode_buf.h

@@ -47,6 +47,7 @@ struct xfs_icdinode {
 	__uint16_t	di_flags;	/* random flags, XFS_DIFLAG_... */
 
 	__uint64_t	di_flags2;	/* more random flags */
+	__uint32_t	di_cowextsize;	/* basic cow extent size for file */
 
 	xfs_ictimestamp_t di_crtime;	/* time created */
 };

fs/xfs/libxfs/xfs_inode_fork.c

@@ -121,6 +121,26 @@ xfs_iformat_fork(
 		return -EFSCORRUPTED;
 	}
 
+	if (unlikely(xfs_is_reflink_inode(ip) &&
+	    (VFS_I(ip)->i_mode & S_IFMT) != S_IFREG)) {
+		xfs_warn(ip->i_mount,
+			"corrupt dinode %llu, wrong file type for reflink.",
+			ip->i_ino);
+		XFS_CORRUPTION_ERROR("xfs_iformat(reflink)",
+				     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
+		return -EFSCORRUPTED;
+	}
+
+	if (unlikely(xfs_is_reflink_inode(ip) &&
+	    (ip->i_d.di_flags & XFS_DIFLAG_REALTIME))) {
+		xfs_warn(ip->i_mount,
+			"corrupt dinode %llu, has reflink+realtime flag set.",
+			ip->i_ino);
+		XFS_CORRUPTION_ERROR("xfs_iformat(reflink)",
+				     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
+		return -EFSCORRUPTED;
+	}
+
 	switch (VFS_I(ip)->i_mode & S_IFMT) {
 	case S_IFIFO:
 	case S_IFCHR:
@@ -186,9 +206,14 @@ xfs_iformat_fork(
 		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
 		return -EFSCORRUPTED;
 	}
-	if (error) {
+	if (error)
 		return error;
+
+	if (xfs_is_reflink_inode(ip)) {
+		ASSERT(ip->i_cowfp == NULL);
+		xfs_ifork_init_cow(ip);
 	}
+
 	if (!XFS_DFORK_Q(dip))
 		return 0;
 
@@ -208,7 +233,8 @@ xfs_iformat_fork(
 			XFS_CORRUPTION_ERROR("xfs_iformat(8)",
 					     XFS_ERRLEVEL_LOW,
 					     ip->i_mount, dip);
-			return -EFSCORRUPTED;
+			error = -EFSCORRUPTED;
+			break;
 		}
 
 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
@@ -226,6 +252,9 @@ xfs_iformat_fork(
 	if (error) {
 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
 		ip->i_afp = NULL;
+		if (ip->i_cowfp)
+			kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
+		ip->i_cowfp = NULL;
 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
 	}
 	return error;
@@ -740,6 +769,9 @@ xfs_idestroy_fork(
 	if (whichfork == XFS_ATTR_FORK) {
 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
 		ip->i_afp = NULL;
+	} else if (whichfork == XFS_COW_FORK) {
+		kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
+		ip->i_cowfp = NULL;
 	}
 }
 
@@ -927,6 +959,19 @@ xfs_iext_get_ext(
 	}
 }
 
+/* Convert bmap state flags to an inode fork. */
+struct xfs_ifork *
+xfs_iext_state_to_fork(
+	struct xfs_inode	*ip,
+	int			state)
+{
+	if (state & BMAP_COWFORK)
+		return ip->i_cowfp;
+	else if (state & BMAP_ATTRFORK)
+		return ip->i_afp;
+	return &ip->i_df;
+}
+
 /*
  * Insert new item(s) into the extent records for incore inode
  * fork 'ifp'.  'count' new items are inserted at index 'idx'.
@@ -939,7 +984,7 @@ xfs_iext_insert(
 	xfs_bmbt_irec_t	*new,		/* items to insert */
 	int		state)		/* type of extent conversion */
 {
-	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
+	xfs_ifork_t	*ifp = xfs_iext_state_to_fork(ip, state);
 	xfs_extnum_t	i;		/* extent record index */
 
 	trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
@@ -1189,7 +1234,7 @@ xfs_iext_remove(
 	int		ext_diff,	/* number of extents to remove */
 	int		state)		/* type of extent conversion */
 {
-	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
+	xfs_ifork_t	*ifp = xfs_iext_state_to_fork(ip, state);
 	xfs_extnum_t	nextents;	/* number of extents in file */
 	int		new_size;	/* size of extents after removal */
 
@@ -1934,3 +1979,20 @@ xfs_iext_irec_update_extoffs(
 		ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
 	}
 }
+
+/*
+ * Initialize an inode's copy-on-write fork.
+ */
+void
+xfs_ifork_init_cow(
+	struct xfs_inode	*ip)
+{
+	if (ip->i_cowfp)
+		return;
+
+	ip->i_cowfp = kmem_zone_zalloc(xfs_ifork_zone,
+				       KM_SLEEP | KM_NOFS);
+	ip->i_cowfp->if_flags = XFS_IFEXTENTS;
+	ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
+	ip->i_cnextents = 0;
+}

fs/xfs/libxfs/xfs_inode_fork.h

@@ -92,7 +92,9 @@ typedef struct xfs_ifork {
 #define XFS_IFORK_PTR(ip,w)		\
 	((w) == XFS_DATA_FORK ? \
 		&(ip)->i_df : \
-		(ip)->i_afp)
+		((w) == XFS_ATTR_FORK ? \
+			(ip)->i_afp : \
+			(ip)->i_cowfp))
 #define XFS_IFORK_DSIZE(ip) \
 	(XFS_IFORK_Q(ip) ? \
 		XFS_IFORK_BOFF(ip) : \
@@ -105,26 +107,38 @@ typedef struct xfs_ifork {
 #define XFS_IFORK_SIZE(ip,w) \
 	((w) == XFS_DATA_FORK ? \
 		XFS_IFORK_DSIZE(ip) : \
-		XFS_IFORK_ASIZE(ip))
+		((w) == XFS_ATTR_FORK ? \
+			XFS_IFORK_ASIZE(ip) : \
+			0))
 #define XFS_IFORK_FORMAT(ip,w) \
 	((w) == XFS_DATA_FORK ? \
 		(ip)->i_d.di_format : \
-		(ip)->i_d.di_aformat)
+		((w) == XFS_ATTR_FORK ? \
+			(ip)->i_d.di_aformat : \
+			(ip)->i_cformat))
 #define XFS_IFORK_FMT_SET(ip,w,n) \
 	((w) == XFS_DATA_FORK ? \
 		((ip)->i_d.di_format = (n)) : \
-		((ip)->i_d.di_aformat = (n)))
+		((w) == XFS_ATTR_FORK ? \
+			((ip)->i_d.di_aformat = (n)) : \
+			((ip)->i_cformat = (n))))
 #define XFS_IFORK_NEXTENTS(ip,w) \
 	((w) == XFS_DATA_FORK ? \
 		(ip)->i_d.di_nextents : \
-		(ip)->i_d.di_anextents)
+		((w) == XFS_ATTR_FORK ? \
+			(ip)->i_d.di_anextents : \
+			(ip)->i_cnextents))
 #define XFS_IFORK_NEXT_SET(ip,w,n) \
 	((w) == XFS_DATA_FORK ? \
 		((ip)->i_d.di_nextents = (n)) : \
-		((ip)->i_d.di_anextents = (n)))
+		((w) == XFS_ATTR_FORK ? \
+			((ip)->i_d.di_anextents = (n)) : \
+			((ip)->i_cnextents = (n))))
 #define XFS_IFORK_MAXEXT(ip, w) \
 	(XFS_IFORK_SIZE(ip, w) / sizeof(xfs_bmbt_rec_t))
 
+struct xfs_ifork *xfs_iext_state_to_fork(struct xfs_inode *ip, int state);
+
 int		xfs_iformat_fork(struct xfs_inode *, struct xfs_dinode *);
 void		xfs_iflush_fork(struct xfs_inode *, struct xfs_dinode *,
 				struct xfs_inode_log_item *, int);
@@ -169,4 +183,6 @@ void		xfs_iext_irec_update_extoffs(struct xfs_ifork *, int, int);
 
 extern struct kmem_zone	*xfs_ifork_zone;
 
+extern void xfs_ifork_init_cow(struct xfs_inode *ip);
+
 #endif	/* __XFS_INODE_FORK_H__ */

fs/xfs/libxfs/xfs_log_format.h

@@ -112,7 +112,11 @@ static inline uint xlog_get_cycle(char *ptr)
 #define XLOG_REG_TYPE_ICREATE		20
 #define XLOG_REG_TYPE_RUI_FORMAT	21
 #define XLOG_REG_TYPE_RUD_FORMAT	22
-#define XLOG_REG_TYPE_MAX		22
+#define XLOG_REG_TYPE_CUI_FORMAT	23
+#define XLOG_REG_TYPE_CUD_FORMAT	24
+#define XLOG_REG_TYPE_BUI_FORMAT	25
+#define XLOG_REG_TYPE_BUD_FORMAT	26
+#define XLOG_REG_TYPE_MAX		26
 
 /*
  * Flags to log operation header
@@ -231,6 +235,10 @@ typedef struct xfs_trans_header {
 #define	XFS_LI_ICREATE		0x123f
 #define	XFS_LI_RUI		0x1240	/* rmap update intent */
 #define	XFS_LI_RUD		0x1241
+#define	XFS_LI_CUI		0x1242	/* refcount update intent */
+#define	XFS_LI_CUD		0x1243
+#define	XFS_LI_BUI		0x1244	/* bmbt update intent */
+#define	XFS_LI_BUD		0x1245
 
 #define XFS_LI_TYPE_DESC \
 	{ XFS_LI_EFI,		"XFS_LI_EFI" }, \
@@ -242,7 +250,11 @@ typedef struct xfs_trans_header {
 	{ XFS_LI_QUOTAOFF,	"XFS_LI_QUOTAOFF" }, \
 	{ XFS_LI_ICREATE,	"XFS_LI_ICREATE" }, \
 	{ XFS_LI_RUI,		"XFS_LI_RUI" }, \
-	{ XFS_LI_RUD,		"XFS_LI_RUD" }
+	{ XFS_LI_RUD,		"XFS_LI_RUD" }, \
+	{ XFS_LI_CUI,		"XFS_LI_CUI" }, \
+	{ XFS_LI_CUD,		"XFS_LI_CUD" }, \
+	{ XFS_LI_BUI,		"XFS_LI_BUI" }, \
+	{ XFS_LI_BUD,		"XFS_LI_BUD" }
 
 /*
  * Inode Log Item Format definitions.
@@ -411,7 +423,8 @@ struct xfs_log_dinode {
 	__uint64_t	di_changecount;	/* number of attribute changes */
 	xfs_lsn_t	di_lsn;		/* flush sequence */
 	__uint64_t	di_flags2;	/* more random flags */
-	__uint8_t	di_pad2[16];	/* more padding for future expansion */
+	__uint32_t	di_cowextsize;	/* basic cow extent size for file */
+	__uint8_t	di_pad2[12];	/* more padding for future expansion */
 
 	/* fields only written to during inode creation */
 	xfs_ictimestamp_t di_crtime;	/* time created */
@@ -622,8 +635,11 @@ struct xfs_map_extent {
 
 /* rmap me_flags: upper bits are flags, lower byte is type code */
 #define XFS_RMAP_EXTENT_MAP		1
+#define XFS_RMAP_EXTENT_MAP_SHARED	2
 #define XFS_RMAP_EXTENT_UNMAP		3
+#define XFS_RMAP_EXTENT_UNMAP_SHARED	4
 #define XFS_RMAP_EXTENT_CONVERT		5
+#define XFS_RMAP_EXTENT_CONVERT_SHARED	6
 #define XFS_RMAP_EXTENT_ALLOC		7
 #define XFS_RMAP_EXTENT_FREE		8
 #define XFS_RMAP_EXTENT_TYPE_MASK	0xFF
@@ -670,6 +686,102 @@ struct xfs_rud_log_format {
 	__uint64_t		rud_rui_id;	/* id of corresponding rui */
 };
 
+/*
+ * CUI/CUD (refcount update) log format definitions
+ */
+struct xfs_phys_extent {
+	__uint64_t		pe_startblock;
+	__uint32_t		pe_len;
+	__uint32_t		pe_flags;
+};
+
+/* refcount pe_flags: upper bits are flags, lower byte is type code */
+/* Type codes are taken directly from enum xfs_refcount_intent_type. */
+#define XFS_REFCOUNT_EXTENT_TYPE_MASK	0xFF
+
+#define XFS_REFCOUNT_EXTENT_FLAGS	(XFS_REFCOUNT_EXTENT_TYPE_MASK)
+
+/*
+ * This is the structure used to lay out a cui log item in the
+ * log.  The cui_extents field is a variable size array whose
+ * size is given by cui_nextents.
+ */
+struct xfs_cui_log_format {
+	__uint16_t		cui_type;	/* cui log item type */
+	__uint16_t		cui_size;	/* size of this item */
+	__uint32_t		cui_nextents;	/* # extents to free */
+	__uint64_t		cui_id;		/* cui identifier */
+	struct xfs_phys_extent	cui_extents[];	/* array of extents */
+};
+
+static inline size_t
+xfs_cui_log_format_sizeof(
+	unsigned int		nr)
+{
+	return sizeof(struct xfs_cui_log_format) +
+			nr * sizeof(struct xfs_phys_extent);
+}
+
+/*
+ * This is the structure used to lay out a cud log item in the
+ * log.  The cud_extents array is a variable size array whose
+ * size is given by cud_nextents;
+ */
+struct xfs_cud_log_format {
+	__uint16_t		cud_type;	/* cud log item type */
+	__uint16_t		cud_size;	/* size of this item */
+	__uint32_t		__pad;
+	__uint64_t		cud_cui_id;	/* id of corresponding cui */
+};
+
+/*
+ * BUI/BUD (inode block mapping) log format definitions
+ */
+
+/* bmbt me_flags: upper bits are flags, lower byte is type code */
+/* Type codes are taken directly from enum xfs_bmap_intent_type. */
+#define XFS_BMAP_EXTENT_TYPE_MASK	0xFF
+
+#define XFS_BMAP_EXTENT_ATTR_FORK	(1U << 31)
+#define XFS_BMAP_EXTENT_UNWRITTEN	(1U << 30)
+
+#define XFS_BMAP_EXTENT_FLAGS		(XFS_BMAP_EXTENT_TYPE_MASK | \
+					 XFS_BMAP_EXTENT_ATTR_FORK | \
+					 XFS_BMAP_EXTENT_UNWRITTEN)
+
+/*
+ * This is the structure used to lay out an bui log item in the
+ * log.  The bui_extents field is a variable size array whose
+ * size is given by bui_nextents.
+ */
+struct xfs_bui_log_format {
+	__uint16_t		bui_type;	/* bui log item type */
+	__uint16_t		bui_size;	/* size of this item */
+	__uint32_t		bui_nextents;	/* # extents to free */
+	__uint64_t		bui_id;		/* bui identifier */
+	struct xfs_map_extent	bui_extents[];	/* array of extents to bmap */
+};
+
+static inline size_t
+xfs_bui_log_format_sizeof(
+	unsigned int		nr)
+{
+	return sizeof(struct xfs_bui_log_format) +
+			nr * sizeof(struct xfs_map_extent);
+}
+
+/*
+ * This is the structure used to lay out an bud log item in the
+ * log.  The bud_extents array is a variable size array whose
+ * size is given by bud_nextents;
+ */
+struct xfs_bud_log_format {
+	__uint16_t		bud_type;	/* bud log item type */
+	__uint16_t		bud_size;	/* size of this item */
+	__uint32_t		__pad;
+	__uint64_t		bud_bui_id;	/* id of corresponding bui */
+};
+
 /*
  * Dquot Log format definitions.
  *

fs/xfs/libxfs/xfs_refcount.c

@@ -0,0 +1,1698 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_sb.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_btree.h"
+#include "xfs_bmap.h"
+#include "xfs_refcount_btree.h"
+#include "xfs_alloc.h"
+#include "xfs_error.h"
+#include "xfs_trace.h"
+#include "xfs_cksum.h"
+#include "xfs_trans.h"
+#include "xfs_bit.h"
+#include "xfs_refcount.h"
+#include "xfs_rmap.h"
+
+/* Allowable refcount adjustment amounts. */
+enum xfs_refc_adjust_op {
+	XFS_REFCOUNT_ADJUST_INCREASE	= 1,
+	XFS_REFCOUNT_ADJUST_DECREASE	= -1,
+	XFS_REFCOUNT_ADJUST_COW_ALLOC	= 0,
+	XFS_REFCOUNT_ADJUST_COW_FREE	= -1,
+};
+
+STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
+		xfs_agblock_t agbno, xfs_extlen_t aglen,
+		struct xfs_defer_ops *dfops);
+STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
+		xfs_agblock_t agbno, xfs_extlen_t aglen,
+		struct xfs_defer_ops *dfops);
+
+/*
+ * Look up the first record less than or equal to [bno, len] in the btree
+ * given by cur.
+ */
+int
+xfs_refcount_lookup_le(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	int			*stat)
+{
+	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
+			XFS_LOOKUP_LE);
+	cur->bc_rec.rc.rc_startblock = bno;
+	cur->bc_rec.rc.rc_blockcount = 0;
+	return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
+}
+
+/*
+ * Look up the first record greater than or equal to [bno, len] in the btree
+ * given by cur.
+ */
+int
+xfs_refcount_lookup_ge(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	int			*stat)
+{
+	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
+			XFS_LOOKUP_GE);
+	cur->bc_rec.rc.rc_startblock = bno;
+	cur->bc_rec.rc.rc_blockcount = 0;
+	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
+}
+
+/* Convert on-disk record to in-core format. */
+static inline void
+xfs_refcount_btrec_to_irec(
+	union xfs_btree_rec		*rec,
+	struct xfs_refcount_irec	*irec)
+{
+	irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
+	irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
+	irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
+}
+
+/*
+ * Get the data from the pointed-to record.
+ */
+int
+xfs_refcount_get_rec(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*irec,
+	int				*stat)
+{
+	union xfs_btree_rec		*rec;
+	int				error;
+
+	error = xfs_btree_get_rec(cur, &rec, stat);
+	if (!error && *stat == 1) {
+		xfs_refcount_btrec_to_irec(rec, irec);
+		trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno,
+				irec);
+	}
+	return error;
+}
+
+/*
+ * Update the record referred to by cur to the value given
+ * by [bno, len, refcount].
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_update(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*irec)
+{
+	union xfs_btree_rec	rec;
+	int			error;
+
+	trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
+	rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
+	rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
+	rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
+	error = xfs_btree_update(cur, &rec);
+	if (error)
+		trace_xfs_refcount_update_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Insert the record referred to by cur to the value given
+ * by [bno, len, refcount].
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_insert(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*irec,
+	int				*i)
+{
+	int				error;
+
+	trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
+	cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
+	cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
+	cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
+	error = xfs_btree_insert(cur, i);
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
+out_error:
+	if (error)
+		trace_xfs_refcount_insert_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Remove the record referred to by cur, then set the pointer to the spot
+ * where the record could be re-inserted, in case we want to increment or
+ * decrement the cursor.
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_delete(
+	struct xfs_btree_cur	*cur,
+	int			*i)
+{
+	struct xfs_refcount_irec	irec;
+	int			found_rec;
+	int			error;
+
+	error = xfs_refcount_get_rec(cur, &irec, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+	trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
+	error = xfs_btree_delete(cur, i);
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
+	if (error)
+		goto out_error;
+	error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
+out_error:
+	if (error)
+		trace_xfs_refcount_delete_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Adjusting the Reference Count
+ *
+ * As stated elsewhere, the reference count btree (refcbt) stores
+ * >1 reference counts for extents of physical blocks.  In this
+ * operation, we're either raising or lowering the reference count of
+ * some subrange stored in the tree:
+ *
+ *      <------ adjustment range ------>
+ * ----+   +---+-----+ +--+--------+---------
+ *  2  |   | 3 |  4  | |17|   55   |   10
+ * ----+   +---+-----+ +--+--------+---------
+ * X axis is physical blocks number;
+ * reference counts are the numbers inside the rectangles
+ *
+ * The first thing we need to do is to ensure that there are no
+ * refcount extents crossing either boundary of the range to be
+ * adjusted.  For any extent that does cross a boundary, split it into
+ * two extents so that we can increment the refcount of one of the
+ * pieces later:
+ *
+ *      <------ adjustment range ------>
+ * ----+   +---+-----+ +--+--------+----+----
+ *  2  |   | 3 |  2  | |17|   55   | 10 | 10
+ * ----+   +---+-----+ +--+--------+----+----
+ *
+ * For this next step, let's assume that all the physical blocks in
+ * the adjustment range are mapped to a file and are therefore in use
+ * at least once.  Therefore, we can infer that any gap in the
+ * refcount tree within the adjustment range represents a physical
+ * extent with refcount == 1:
+ *
+ *      <------ adjustment range ------>
+ * ----+---+---+-----+-+--+--------+----+----
+ *  2  |"1"| 3 |  2  |1|17|   55   | 10 | 10
+ * ----+---+---+-----+-+--+--------+----+----
+ *      ^
+ *
+ * For each extent that falls within the interval range, figure out
+ * which extent is to the left or the right of that extent.  Now we
+ * have a left, current, and right extent.  If the new reference count
+ * of the center extent enables us to merge left, center, and right
+ * into one record covering all three, do so.  If the center extent is
+ * at the left end of the range, abuts the left extent, and its new
+ * reference count matches the left extent's record, then merge them.
+ * If the center extent is at the right end of the range, abuts the
+ * right extent, and the reference counts match, merge those.  In the
+ * example, we can left merge (assuming an increment operation):
+ *
+ *      <------ adjustment range ------>
+ * --------+---+-----+-+--+--------+----+----
+ *    2    | 3 |  2  |1|17|   55   | 10 | 10
+ * --------+---+-----+-+--+--------+----+----
+ *          ^
+ *
+ * For all other extents within the range, adjust the reference count
+ * or delete it if the refcount falls below 2.  If we were
+ * incrementing, the end result looks like this:
+ *
+ *      <------ adjustment range ------>
+ * --------+---+-----+-+--+--------+----+----
+ *    2    | 4 |  3  |2|18|   56   | 11 | 10
+ * --------+---+-----+-+--+--------+----+----
+ *
+ * The result of a decrement operation looks as such:
+ *
+ *      <------ adjustment range ------>
+ * ----+   +---+       +--+--------+----+----
+ *  2  |   | 2 |       |16|   54   |  9 | 10
+ * ----+   +---+       +--+--------+----+----
+ *      DDDD    111111DD
+ *
+ * The blocks marked "D" are freed; the blocks marked "1" are only
+ * referenced once and therefore the record is removed from the
+ * refcount btree.
+ */
+
+/* Next block after this extent. */
+static inline xfs_agblock_t
+xfs_refc_next(
+	struct xfs_refcount_irec	*rc)
+{
+	return rc->rc_startblock + rc->rc_blockcount;
+}
+
+/*
+ * Split a refcount extent that crosses agbno.
+ */
+STATIC int
+xfs_refcount_split_extent(
+	struct xfs_btree_cur		*cur,
+	xfs_agblock_t			agbno,
+	bool				*shape_changed)
+{
+	struct xfs_refcount_irec	rcext, tmp;
+	int				found_rec;
+	int				error;
+
+	*shape_changed = false;
+	error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
+	if (error)
+		goto out_error;
+	if (!found_rec)
+		return 0;
+
+	error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+	if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
+		return 0;
+
+	*shape_changed = true;
+	trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_private.a.agno,
+			&rcext, agbno);
+
+	/* Establish the right extent. */
+	tmp = rcext;
+	tmp.rc_startblock = agbno;
+	tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
+	error = xfs_refcount_update(cur, &tmp);
+	if (error)
+		goto out_error;
+
+	/* Insert the left extent. */
+	tmp = rcext;
+	tmp.rc_blockcount = agbno - rcext.rc_startblock;
+	error = xfs_refcount_insert(cur, &tmp, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+	return error;
+
+out_error:
+	trace_xfs_refcount_split_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Merge the left, center, and right extents.
+ */
+STATIC int
+xfs_refcount_merge_center_extents(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*left,
+	struct xfs_refcount_irec	*center,
+	struct xfs_refcount_irec	*right,
+	unsigned long long		extlen,
+	xfs_agblock_t			*agbno,
+	xfs_extlen_t			*aglen)
+{
+	int				error;
+	int				found_rec;
+
+	trace_xfs_refcount_merge_center_extents(cur->bc_mp,
+			cur->bc_private.a.agno, left, center, right);
+
+	/*
+	 * Make sure the center and right extents are not in the btree.
+	 * If the center extent was synthesized, the first delete call
+	 * removes the right extent and we skip the second deletion.
+	 * If center and right were in the btree, then the first delete
+	 * call removes the center and the second one removes the right
+	 * extent.
+	 */
+	error = xfs_refcount_lookup_ge(cur, center->rc_startblock,
+			&found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	error = xfs_refcount_delete(cur, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	if (center->rc_refcount > 1) {
+		error = xfs_refcount_delete(cur, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+	}
+
+	/* Enlarge the left extent. */
+	error = xfs_refcount_lookup_le(cur, left->rc_startblock,
+			&found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	left->rc_blockcount = extlen;
+	error = xfs_refcount_update(cur, left);
+	if (error)
+		goto out_error;
+
+	*aglen = 0;
+	return error;
+
+out_error:
+	trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Merge with the left extent.
+ */
+STATIC int
+xfs_refcount_merge_left_extent(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*left,
+	struct xfs_refcount_irec	*cleft,
+	xfs_agblock_t			*agbno,
+	xfs_extlen_t			*aglen)
+{
+	int				error;
+	int				found_rec;
+
+	trace_xfs_refcount_merge_left_extent(cur->bc_mp,
+			cur->bc_private.a.agno, left, cleft);
+
+	/* If the extent at agbno (cleft) wasn't synthesized, remove it. */
+	if (cleft->rc_refcount > 1) {
+		error = xfs_refcount_lookup_le(cur, cleft->rc_startblock,
+				&found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+
+		error = xfs_refcount_delete(cur, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+	}
+
+	/* Enlarge the left extent. */
+	error = xfs_refcount_lookup_le(cur, left->rc_startblock,
+			&found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	left->rc_blockcount += cleft->rc_blockcount;
+	error = xfs_refcount_update(cur, left);
+	if (error)
+		goto out_error;
+
+	*agbno += cleft->rc_blockcount;
+	*aglen -= cleft->rc_blockcount;
+	return error;
+
+out_error:
+	trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Merge with the right extent.
+ */
+STATIC int
+xfs_refcount_merge_right_extent(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*right,
+	struct xfs_refcount_irec	*cright,
+	xfs_agblock_t			*agbno,
+	xfs_extlen_t			*aglen)
+{
+	int				error;
+	int				found_rec;
+
+	trace_xfs_refcount_merge_right_extent(cur->bc_mp,
+			cur->bc_private.a.agno, cright, right);
+
+	/*
+	 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
+	 * remove it.
+	 */
+	if (cright->rc_refcount > 1) {
+		error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
+			&found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+
+		error = xfs_refcount_delete(cur, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+	}
+
+	/* Enlarge the right extent. */
+	error = xfs_refcount_lookup_le(cur, right->rc_startblock,
+			&found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	right->rc_startblock -= cright->rc_blockcount;
+	right->rc_blockcount += cright->rc_blockcount;
+	error = xfs_refcount_update(cur, right);
+	if (error)
+		goto out_error;
+
+	*aglen -= cright->rc_blockcount;
+	return error;
+
+out_error:
+	trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+#define XFS_FIND_RCEXT_SHARED	1
+#define XFS_FIND_RCEXT_COW	2
+/*
+ * Find the left extent and the one after it (cleft).  This function assumes
+ * that we've already split any extent crossing agbno.
+ */
+STATIC int
+xfs_refcount_find_left_extents(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*left,
+	struct xfs_refcount_irec	*cleft,
+	xfs_agblock_t			agbno,
+	xfs_extlen_t			aglen,
+	int				flags)
+{
+	struct xfs_refcount_irec	tmp;
+	int				error;
+	int				found_rec;
+
+	left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
+	error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
+	if (error)
+		goto out_error;
+	if (!found_rec)
+		return 0;
+
+	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	if (xfs_refc_next(&tmp) != agbno)
+		return 0;
+	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
+		return 0;
+	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
+		return 0;
+	/* We have a left extent; retrieve (or invent) the next right one */
+	*left = tmp;
+
+	error = xfs_btree_increment(cur, 0, &found_rec);
+	if (error)
+		goto out_error;
+	if (found_rec) {
+		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+
+		/* if tmp starts at the end of our range, just use that */
+		if (tmp.rc_startblock == agbno)
+			*cleft = tmp;
+		else {
+			/*
+			 * There's a gap in the refcntbt at the start of the
+			 * range we're interested in (refcount == 1) so
+			 * synthesize the implied extent and pass it back.
+			 * We assume here that the agbno/aglen range was
+			 * passed in from a data fork extent mapping and
+			 * therefore is allocated to exactly one owner.
+			 */
+			cleft->rc_startblock = agbno;
+			cleft->rc_blockcount = min(aglen,
+					tmp.rc_startblock - agbno);
+			cleft->rc_refcount = 1;
+		}
+	} else {
+		/*
+		 * No extents, so pretend that there's one covering the whole
+		 * range.
+		 */
+		cleft->rc_startblock = agbno;
+		cleft->rc_blockcount = aglen;
+		cleft->rc_refcount = 1;
+	}
+	trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_private.a.agno,
+			left, cleft, agbno);
+	return error;
+
+out_error:
+	trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Find the right extent and the one before it (cright).  This function
+ * assumes that we've already split any extents crossing agbno + aglen.
+ */
+STATIC int
+xfs_refcount_find_right_extents(
+	struct xfs_btree_cur		*cur,
+	struct xfs_refcount_irec	*right,
+	struct xfs_refcount_irec	*cright,
+	xfs_agblock_t			agbno,
+	xfs_extlen_t			aglen,
+	int				flags)
+{
+	struct xfs_refcount_irec	tmp;
+	int				error;
+	int				found_rec;
+
+	right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
+	error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
+	if (error)
+		goto out_error;
+	if (!found_rec)
+		return 0;
+
+	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+
+	if (tmp.rc_startblock != agbno + aglen)
+		return 0;
+	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
+		return 0;
+	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
+		return 0;
+	/* We have a right extent; retrieve (or invent) the next left one */
+	*right = tmp;
+
+	error = xfs_btree_decrement(cur, 0, &found_rec);
+	if (error)
+		goto out_error;
+	if (found_rec) {
+		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
+				out_error);
+
+		/* if tmp ends at the end of our range, just use that */
+		if (xfs_refc_next(&tmp) == agbno + aglen)
+			*cright = tmp;
+		else {
+			/*
+			 * There's a gap in the refcntbt at the end of the
+			 * range we're interested in (refcount == 1) so
+			 * create the implied extent and pass it back.
+			 * We assume here that the agbno/aglen range was
+			 * passed in from a data fork extent mapping and
+			 * therefore is allocated to exactly one owner.
+			 */
+			cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
+			cright->rc_blockcount = right->rc_startblock -
+					cright->rc_startblock;
+			cright->rc_refcount = 1;
+		}
+	} else {
+		/*
+		 * No extents, so pretend that there's one covering the whole
+		 * range.
+		 */
+		cright->rc_startblock = agbno;
+		cright->rc_blockcount = aglen;
+		cright->rc_refcount = 1;
+	}
+	trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_private.a.agno,
+			cright, right, agbno + aglen);
+	return error;
+
+out_error:
+	trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/* Is this extent valid? */
+static inline bool
+xfs_refc_valid(
+	struct xfs_refcount_irec	*rc)
+{
+	return rc->rc_startblock != NULLAGBLOCK;
+}
+
+/*
+ * Try to merge with any extents on the boundaries of the adjustment range.
+ */
+STATIC int
+xfs_refcount_merge_extents(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		*agbno,
+	xfs_extlen_t		*aglen,
+	enum xfs_refc_adjust_op adjust,
+	int			flags,
+	bool			*shape_changed)
+{
+	struct xfs_refcount_irec	left = {0}, cleft = {0};
+	struct xfs_refcount_irec	cright = {0}, right = {0};
+	int				error;
+	unsigned long long		ulen;
+	bool				cequal;
+
+	*shape_changed = false;
+	/*
+	 * Find the extent just below agbno [left], just above agbno [cleft],
+	 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
+	 * [right].
+	 */
+	error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
+			*aglen, flags);
+	if (error)
+		return error;
+	error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
+			*aglen, flags);
+	if (error)
+		return error;
+
+	/* No left or right extent to merge; exit. */
+	if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
+		return 0;
+
+	cequal = (cleft.rc_startblock == cright.rc_startblock) &&
+		 (cleft.rc_blockcount == cright.rc_blockcount);
+
+	/* Try to merge left, cleft, and right.  cleft must == cright. */
+	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
+			right.rc_blockcount;
+	if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
+	    xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
+	    left.rc_refcount == cleft.rc_refcount + adjust &&
+	    right.rc_refcount == cleft.rc_refcount + adjust &&
+	    ulen < MAXREFCEXTLEN) {
+		*shape_changed = true;
+		return xfs_refcount_merge_center_extents(cur, &left, &cleft,
+				&right, ulen, agbno, aglen);
+	}
+
+	/* Try to merge left and cleft. */
+	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
+	if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
+	    left.rc_refcount == cleft.rc_refcount + adjust &&
+	    ulen < MAXREFCEXTLEN) {
+		*shape_changed = true;
+		error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
+				agbno, aglen);
+		if (error)
+			return error;
+
+		/*
+		 * If we just merged left + cleft and cleft == cright,
+		 * we no longer have a cright to merge with right.  We're done.
+		 */
+		if (cequal)
+			return 0;
+	}
+
+	/* Try to merge cright and right. */
+	ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
+	if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
+	    right.rc_refcount == cright.rc_refcount + adjust &&
+	    ulen < MAXREFCEXTLEN) {
+		*shape_changed = true;
+		return xfs_refcount_merge_right_extent(cur, &right, &cright,
+				agbno, aglen);
+	}
+
+	return error;
+}
+
+/*
+ * While we're adjusting the refcounts records of an extent, we have
+ * to keep an eye on the number of extents we're dirtying -- run too
+ * many in a single transaction and we'll exceed the transaction's
+ * reservation and crash the fs.  Each record adds 12 bytes to the
+ * log (plus any key updates) so we'll conservatively assume 24 bytes
+ * per record.  We must also leave space for btree splits on both ends
+ * of the range and space for the CUD and a new CUI.
+ *
+ * XXX: This is a pretty hand-wavy estimate.  The penalty for guessing
+ * true incorrectly is a shutdown FS; the penalty for guessing false
+ * incorrectly is more transaction rolls than might be necessary.
+ * Be conservative here.
+ */
+static bool
+xfs_refcount_still_have_space(
+	struct xfs_btree_cur		*cur)
+{
+	unsigned long			overhead;
+
+	overhead = cur->bc_private.a.priv.refc.shape_changes *
+			xfs_allocfree_log_count(cur->bc_mp, 1);
+	overhead *= cur->bc_mp->m_sb.sb_blocksize;
+
+	/*
+	 * Only allow 2 refcount extent updates per transaction if the
+	 * refcount continue update "error" has been injected.
+	 */
+	if (cur->bc_private.a.priv.refc.nr_ops > 2 &&
+	    XFS_TEST_ERROR(false, cur->bc_mp,
+			XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE,
+			XFS_RANDOM_REFCOUNT_CONTINUE_UPDATE))
+		return false;
+
+	if (cur->bc_private.a.priv.refc.nr_ops == 0)
+		return true;
+	else if (overhead > cur->bc_tp->t_log_res)
+		return false;
+	return  cur->bc_tp->t_log_res - overhead >
+		cur->bc_private.a.priv.refc.nr_ops * 32;
+}
+
+/*
+ * Adjust the refcounts of middle extents.  At this point we should have
+ * split extents that crossed the adjustment range; merged with adjacent
+ * extents; and updated agbno/aglen to reflect the merges.  Therefore,
+ * all we have to do is update the extents inside [agbno, agbno + aglen].
+ */
+STATIC int
+xfs_refcount_adjust_extents(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		*agbno,
+	xfs_extlen_t		*aglen,
+	enum xfs_refc_adjust_op	adj,
+	struct xfs_defer_ops	*dfops,
+	struct xfs_owner_info	*oinfo)
+{
+	struct xfs_refcount_irec	ext, tmp;
+	int				error;
+	int				found_rec, found_tmp;
+	xfs_fsblock_t			fsbno;
+
+	/* Merging did all the work already. */
+	if (*aglen == 0)
+		return 0;
+
+	error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
+	if (error)
+		goto out_error;
+
+	while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
+		error = xfs_refcount_get_rec(cur, &ext, &found_rec);
+		if (error)
+			goto out_error;
+		if (!found_rec) {
+			ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
+			ext.rc_blockcount = 0;
+			ext.rc_refcount = 0;
+		}
+
+		/*
+		 * Deal with a hole in the refcount tree; if a file maps to
+		 * these blocks and there's no refcountbt record, pretend that
+		 * there is one with refcount == 1.
+		 */
+		if (ext.rc_startblock != *agbno) {
+			tmp.rc_startblock = *agbno;
+			tmp.rc_blockcount = min(*aglen,
+					ext.rc_startblock - *agbno);
+			tmp.rc_refcount = 1 + adj;
+			trace_xfs_refcount_modify_extent(cur->bc_mp,
+					cur->bc_private.a.agno, &tmp);
+
+			/*
+			 * Either cover the hole (increment) or
+			 * delete the range (decrement).
+			 */
+			if (tmp.rc_refcount) {
+				error = xfs_refcount_insert(cur, &tmp,
+						&found_tmp);
+				if (error)
+					goto out_error;
+				XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+						found_tmp == 1, out_error);
+				cur->bc_private.a.priv.refc.nr_ops++;
+			} else {
+				fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
+						cur->bc_private.a.agno,
+						tmp.rc_startblock);
+				xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
+						tmp.rc_blockcount, oinfo);
+			}
+
+			(*agbno) += tmp.rc_blockcount;
+			(*aglen) -= tmp.rc_blockcount;
+
+			error = xfs_refcount_lookup_ge(cur, *agbno,
+					&found_rec);
+			if (error)
+				goto out_error;
+		}
+
+		/* Stop if there's nothing left to modify */
+		if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
+			break;
+
+		/*
+		 * Adjust the reference count and either update the tree
+		 * (incr) or free the blocks (decr).
+		 */
+		if (ext.rc_refcount == MAXREFCOUNT)
+			goto skip;
+		ext.rc_refcount += adj;
+		trace_xfs_refcount_modify_extent(cur->bc_mp,
+				cur->bc_private.a.agno, &ext);
+		if (ext.rc_refcount > 1) {
+			error = xfs_refcount_update(cur, &ext);
+			if (error)
+				goto out_error;
+			cur->bc_private.a.priv.refc.nr_ops++;
+		} else if (ext.rc_refcount == 1) {
+			error = xfs_refcount_delete(cur, &found_rec);
+			if (error)
+				goto out_error;
+			XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+					found_rec == 1, out_error);
+			cur->bc_private.a.priv.refc.nr_ops++;
+			goto advloop;
+		} else {
+			fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
+					cur->bc_private.a.agno,
+					ext.rc_startblock);
+			xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
+					ext.rc_blockcount, oinfo);
+		}
+
+skip:
+		error = xfs_btree_increment(cur, 0, &found_rec);
+		if (error)
+			goto out_error;
+
+advloop:
+		(*agbno) += ext.rc_blockcount;
+		(*aglen) -= ext.rc_blockcount;
+	}
+
+	return error;
+out_error:
+	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/* Adjust the reference count of a range of AG blocks. */
+STATIC int
+xfs_refcount_adjust(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	xfs_agblock_t		*new_agbno,
+	xfs_extlen_t		*new_aglen,
+	enum xfs_refc_adjust_op	adj,
+	struct xfs_defer_ops	*dfops,
+	struct xfs_owner_info	*oinfo)
+{
+	bool			shape_changed;
+	int			shape_changes = 0;
+	int			error;
+
+	*new_agbno = agbno;
+	*new_aglen = aglen;
+	if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
+		trace_xfs_refcount_increase(cur->bc_mp, cur->bc_private.a.agno,
+				agbno, aglen);
+	else
+		trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_private.a.agno,
+				agbno, aglen);
+
+	/*
+	 * Ensure that no rcextents cross the boundary of the adjustment range.
+	 */
+	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
+	if (error)
+		goto out_error;
+	if (shape_changed)
+		shape_changes++;
+
+	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
+	if (error)
+		goto out_error;
+	if (shape_changed)
+		shape_changes++;
+
+	/*
+	 * Try to merge with the left or right extents of the range.
+	 */
+	error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
+			XFS_FIND_RCEXT_SHARED, &shape_changed);
+	if (error)
+		goto out_error;
+	if (shape_changed)
+		shape_changes++;
+	if (shape_changes)
+		cur->bc_private.a.priv.refc.shape_changes++;
+
+	/* Now that we've taken care of the ends, adjust the middle extents */
+	error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
+			adj, dfops, oinfo);
+	if (error)
+		goto out_error;
+
+	return 0;
+
+out_error:
+	trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_private.a.agno,
+			error, _RET_IP_);
+	return error;
+}
+
+/* Clean up after calling xfs_refcount_finish_one. */
+void
+xfs_refcount_finish_one_cleanup(
+	struct xfs_trans	*tp,
+	struct xfs_btree_cur	*rcur,
+	int			error)
+{
+	struct xfs_buf		*agbp;
+
+	if (rcur == NULL)
+		return;
+	agbp = rcur->bc_private.a.agbp;
+	xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
+	if (error)
+		xfs_trans_brelse(tp, agbp);
+}
+
+/*
+ * Process one of the deferred refcount operations.  We pass back the
+ * btree cursor to maintain our lock on the btree between calls.
+ * This saves time and eliminates a buffer deadlock between the
+ * superblock and the AGF because we'll always grab them in the same
+ * order.
+ */
+int
+xfs_refcount_finish_one(
+	struct xfs_trans		*tp,
+	struct xfs_defer_ops		*dfops,
+	enum xfs_refcount_intent_type	type,
+	xfs_fsblock_t			startblock,
+	xfs_extlen_t			blockcount,
+	xfs_fsblock_t			*new_fsb,
+	xfs_extlen_t			*new_len,
+	struct xfs_btree_cur		**pcur)
+{
+	struct xfs_mount		*mp = tp->t_mountp;
+	struct xfs_btree_cur		*rcur;
+	struct xfs_buf			*agbp = NULL;
+	int				error = 0;
+	xfs_agnumber_t			agno;
+	xfs_agblock_t			bno;
+	xfs_agblock_t			new_agbno;
+	unsigned long			nr_ops = 0;
+	int				shape_changes = 0;
+
+	agno = XFS_FSB_TO_AGNO(mp, startblock);
+	ASSERT(agno != NULLAGNUMBER);
+	bno = XFS_FSB_TO_AGBNO(mp, startblock);
+
+	trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
+			type, XFS_FSB_TO_AGBNO(mp, startblock),
+			blockcount);
+
+	if (XFS_TEST_ERROR(false, mp,
+			XFS_ERRTAG_REFCOUNT_FINISH_ONE,
+			XFS_RANDOM_REFCOUNT_FINISH_ONE))
+		return -EIO;
+
+	/*
+	 * If we haven't gotten a cursor or the cursor AG doesn't match
+	 * the startblock, get one now.
+	 */
+	rcur = *pcur;
+	if (rcur != NULL && rcur->bc_private.a.agno != agno) {
+		nr_ops = rcur->bc_private.a.priv.refc.nr_ops;
+		shape_changes = rcur->bc_private.a.priv.refc.shape_changes;
+		xfs_refcount_finish_one_cleanup(tp, rcur, 0);
+		rcur = NULL;
+		*pcur = NULL;
+	}
+	if (rcur == NULL) {
+		error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
+				XFS_ALLOC_FLAG_FREEING, &agbp);
+		if (error)
+			return error;
+		if (!agbp)
+			return -EFSCORRUPTED;
+
+		rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, dfops);
+		if (!rcur) {
+			error = -ENOMEM;
+			goto out_cur;
+		}
+		rcur->bc_private.a.priv.refc.nr_ops = nr_ops;
+		rcur->bc_private.a.priv.refc.shape_changes = shape_changes;
+	}
+	*pcur = rcur;
+
+	switch (type) {
+	case XFS_REFCOUNT_INCREASE:
+		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
+			new_len, XFS_REFCOUNT_ADJUST_INCREASE, dfops, NULL);
+		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
+		break;
+	case XFS_REFCOUNT_DECREASE:
+		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
+			new_len, XFS_REFCOUNT_ADJUST_DECREASE, dfops, NULL);
+		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
+		break;
+	case XFS_REFCOUNT_ALLOC_COW:
+		*new_fsb = startblock + blockcount;
+		*new_len = 0;
+		error = __xfs_refcount_cow_alloc(rcur, bno, blockcount, dfops);
+		break;
+	case XFS_REFCOUNT_FREE_COW:
+		*new_fsb = startblock + blockcount;
+		*new_len = 0;
+		error = __xfs_refcount_cow_free(rcur, bno, blockcount, dfops);
+		break;
+	default:
+		ASSERT(0);
+		error = -EFSCORRUPTED;
+	}
+	if (!error && *new_len > 0)
+		trace_xfs_refcount_finish_one_leftover(mp, agno, type,
+				bno, blockcount, new_agbno, *new_len);
+	return error;
+
+out_cur:
+	xfs_trans_brelse(tp, agbp);
+
+	return error;
+}
+
+/*
+ * Record a refcount intent for later processing.
+ */
+static int
+__xfs_refcount_add(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	enum xfs_refcount_intent_type	type,
+	xfs_fsblock_t			startblock,
+	xfs_extlen_t			blockcount)
+{
+	struct xfs_refcount_intent	*ri;
+
+	trace_xfs_refcount_defer(mp, XFS_FSB_TO_AGNO(mp, startblock),
+			type, XFS_FSB_TO_AGBNO(mp, startblock),
+			blockcount);
+
+	ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
+			KM_SLEEP | KM_NOFS);
+	INIT_LIST_HEAD(&ri->ri_list);
+	ri->ri_type = type;
+	ri->ri_startblock = startblock;
+	ri->ri_blockcount = blockcount;
+
+	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
+	return 0;
+}
+
+/*
+ * Increase the reference count of the blocks backing a file's extent.
+ */
+int
+xfs_refcount_increase_extent(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	struct xfs_bmbt_irec		*PREV)
+{
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_INCREASE,
+			PREV->br_startblock, PREV->br_blockcount);
+}
+
+/*
+ * Decrease the reference count of the blocks backing a file's extent.
+ */
+int
+xfs_refcount_decrease_extent(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	struct xfs_bmbt_irec		*PREV)
+{
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_DECREASE,
+			PREV->br_startblock, PREV->br_blockcount);
+}
+
+/*
+ * 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 set, return the longest contiguous extent of
+ * shared blocks; if not, just return the first extent we find.  If no
+ * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
+ * and 0, respectively.
+ */
+int
+xfs_refcount_find_shared(
+	struct xfs_btree_cur		*cur,
+	xfs_agblock_t			agbno,
+	xfs_extlen_t			aglen,
+	xfs_agblock_t			*fbno,
+	xfs_extlen_t			*flen,
+	bool				find_end_of_shared)
+{
+	struct xfs_refcount_irec	tmp;
+	int				i;
+	int				have;
+	int				error;
+
+	trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_private.a.agno,
+			agbno, aglen);
+
+	/* By default, skip the whole range */
+	*fbno = NULLAGBLOCK;
+	*flen = 0;
+
+	/* Try to find a refcount extent that crosses the start */
+	error = xfs_refcount_lookup_le(cur, agbno, &have);
+	if (error)
+		goto out_error;
+	if (!have) {
+		/* No left extent, look at the next one */
+		error = xfs_btree_increment(cur, 0, &have);
+		if (error)
+			goto out_error;
+		if (!have)
+			goto done;
+	}
+	error = xfs_refcount_get_rec(cur, &tmp, &i);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
+
+	/* If the extent ends before the start, look at the next one */
+	if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
+		error = xfs_btree_increment(cur, 0, &have);
+		if (error)
+			goto out_error;
+		if (!have)
+			goto done;
+		error = xfs_refcount_get_rec(cur, &tmp, &i);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
+	}
+
+	/* If the extent starts after the range we want, bail out */
+	if (tmp.rc_startblock >= agbno + aglen)
+		goto done;
+
+	/* We found the start of a shared extent! */
+	if (tmp.rc_startblock < agbno) {
+		tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
+		tmp.rc_startblock = agbno;
+	}
+
+	*fbno = tmp.rc_startblock;
+	*flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
+	if (!find_end_of_shared)
+		goto done;
+
+	/* Otherwise, find the end of this shared extent */
+	while (*fbno + *flen < agbno + aglen) {
+		error = xfs_btree_increment(cur, 0, &have);
+		if (error)
+			goto out_error;
+		if (!have)
+			break;
+		error = xfs_refcount_get_rec(cur, &tmp, &i);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
+		if (tmp.rc_startblock >= agbno + aglen ||
+		    tmp.rc_startblock != *fbno + *flen)
+			break;
+		*flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
+	}
+
+done:
+	trace_xfs_refcount_find_shared_result(cur->bc_mp,
+			cur->bc_private.a.agno, *fbno, *flen);
+
+out_error:
+	if (error)
+		trace_xfs_refcount_find_shared_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Recovering CoW Blocks After a Crash
+ *
+ * Due to the way that the copy on write mechanism works, there's a window of
+ * opportunity in which we can lose track of allocated blocks during a crash.
+ * Because CoW uses delayed allocation in the in-core CoW fork, writeback
+ * causes blocks to be allocated and stored in the CoW fork.  The blocks are
+ * no longer in the free space btree but are not otherwise recorded anywhere
+ * until the write completes and the blocks are mapped into the file.  A crash
+ * in between allocation and remapping results in the replacement blocks being
+ * lost.  This situation is exacerbated by the CoW extent size hint because
+ * allocations can hang around for long time.
+ *
+ * However, there is a place where we can record these allocations before they
+ * become mappings -- the reference count btree.  The btree does not record
+ * extents with refcount == 1, so we can record allocations with a refcount of
+ * 1.  Blocks being used for CoW writeout cannot be shared, so there should be
+ * no conflict with shared block records.  These mappings should be created
+ * when we allocate blocks to the CoW fork and deleted when they're removed
+ * from the CoW fork.
+ *
+ * Minor nit: records for in-progress CoW allocations and records for shared
+ * extents must never be merged, to preserve the property that (except for CoW
+ * allocations) there are no refcount btree entries with refcount == 1.  The
+ * only time this could potentially happen is when unsharing a block that's
+ * adjacent to CoW allocations, so we must be careful to avoid this.
+ *
+ * At mount time we recover lost CoW allocations by searching the refcount
+ * btree for these refcount == 1 mappings.  These represent CoW allocations
+ * that were in progress at the time the filesystem went down, so we can free
+ * them to get the space back.
+ *
+ * This mechanism is superior to creating EFIs for unmapped CoW extents for
+ * several reasons -- first, EFIs pin the tail of the log and would have to be
+ * periodically relogged to avoid filling up the log.  Second, CoW completions
+ * will have to file an EFD and create new EFIs for whatever remains in the
+ * CoW fork; this partially takes care of (1) but extent-size reservations
+ * will have to periodically relog even if there's no writeout in progress.
+ * This can happen if the CoW extent size hint is set, which you really want.
+ * Third, EFIs cannot currently be automatically relogged into newer
+ * transactions to advance the log tail.  Fourth, stuffing the log full of
+ * EFIs places an upper bound on the number of CoW allocations that can be
+ * held filesystem-wide at any given time.  Recording them in the refcount
+ * btree doesn't require us to maintain any state in memory and doesn't pin
+ * the log.
+ */
+/*
+ * Adjust the refcounts of CoW allocations.  These allocations are "magic"
+ * in that they're not referenced anywhere else in the filesystem, so we
+ * stash them in the refcount btree with a refcount of 1 until either file
+ * remapping (or CoW cancellation) happens.
+ */
+STATIC int
+xfs_refcount_adjust_cow_extents(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	enum xfs_refc_adjust_op	adj,
+	struct xfs_defer_ops	*dfops,
+	struct xfs_owner_info	*oinfo)
+{
+	struct xfs_refcount_irec	ext, tmp;
+	int				error;
+	int				found_rec, found_tmp;
+
+	if (aglen == 0)
+		return 0;
+
+	/* Find any overlapping refcount records */
+	error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
+	if (error)
+		goto out_error;
+	error = xfs_refcount_get_rec(cur, &ext, &found_rec);
+	if (error)
+		goto out_error;
+	if (!found_rec) {
+		ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
+				XFS_REFC_COW_START;
+		ext.rc_blockcount = 0;
+		ext.rc_refcount = 0;
+	}
+
+	switch (adj) {
+	case XFS_REFCOUNT_ADJUST_COW_ALLOC:
+		/* Adding a CoW reservation, there should be nothing here. */
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+				ext.rc_startblock >= agbno + aglen, out_error);
+
+		tmp.rc_startblock = agbno;
+		tmp.rc_blockcount = aglen;
+		tmp.rc_refcount = 1;
+		trace_xfs_refcount_modify_extent(cur->bc_mp,
+				cur->bc_private.a.agno, &tmp);
+
+		error = xfs_refcount_insert(cur, &tmp,
+				&found_tmp);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+				found_tmp == 1, out_error);
+		break;
+	case XFS_REFCOUNT_ADJUST_COW_FREE:
+		/* Removing a CoW reservation, there should be one extent. */
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+			ext.rc_startblock == agbno, out_error);
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+			ext.rc_blockcount == aglen, out_error);
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+			ext.rc_refcount == 1, out_error);
+
+		ext.rc_refcount = 0;
+		trace_xfs_refcount_modify_extent(cur->bc_mp,
+				cur->bc_private.a.agno, &ext);
+		error = xfs_refcount_delete(cur, &found_rec);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
+				found_rec == 1, out_error);
+		break;
+	default:
+		ASSERT(0);
+	}
+
+	return error;
+out_error:
+	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
+			cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Add or remove refcount btree entries for CoW reservations.
+ */
+STATIC int
+xfs_refcount_adjust_cow(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	enum xfs_refc_adjust_op	adj,
+	struct xfs_defer_ops	*dfops)
+{
+	bool			shape_changed;
+	int			error;
+
+	agbno += XFS_REFC_COW_START;
+
+	/*
+	 * Ensure that no rcextents cross the boundary of the adjustment range.
+	 */
+	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
+	if (error)
+		goto out_error;
+
+	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
+	if (error)
+		goto out_error;
+
+	/*
+	 * Try to merge with the left or right extents of the range.
+	 */
+	error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
+			XFS_FIND_RCEXT_COW, &shape_changed);
+	if (error)
+		goto out_error;
+
+	/* Now that we've taken care of the ends, adjust the middle extents */
+	error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj,
+			dfops, NULL);
+	if (error)
+		goto out_error;
+
+	return 0;
+
+out_error:
+	trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_private.a.agno,
+			error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Record a CoW allocation in the refcount btree.
+ */
+STATIC int
+__xfs_refcount_cow_alloc(
+	struct xfs_btree_cur	*rcur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	struct xfs_defer_ops	*dfops)
+{
+	int			error;
+
+	trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_private.a.agno,
+			agbno, aglen);
+
+	/* Add refcount btree reservation */
+	error = xfs_refcount_adjust_cow(rcur, agbno, aglen,
+			XFS_REFCOUNT_ADJUST_COW_ALLOC, dfops);
+	if (error)
+		return error;
+
+	/* Add rmap entry */
+	if (xfs_sb_version_hasrmapbt(&rcur->bc_mp->m_sb)) {
+		error = xfs_rmap_alloc_extent(rcur->bc_mp, dfops,
+				rcur->bc_private.a.agno,
+				agbno, aglen, XFS_RMAP_OWN_COW);
+		if (error)
+			return error;
+	}
+
+	return error;
+}
+
+/*
+ * Remove a CoW allocation from the refcount btree.
+ */
+STATIC int
+__xfs_refcount_cow_free(
+	struct xfs_btree_cur	*rcur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	struct xfs_defer_ops	*dfops)
+{
+	int			error;
+
+	trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_private.a.agno,
+			agbno, aglen);
+
+	/* Remove refcount btree reservation */
+	error = xfs_refcount_adjust_cow(rcur, agbno, aglen,
+			XFS_REFCOUNT_ADJUST_COW_FREE, dfops);
+	if (error)
+		return error;
+
+	/* Remove rmap entry */
+	if (xfs_sb_version_hasrmapbt(&rcur->bc_mp->m_sb)) {
+		error = xfs_rmap_free_extent(rcur->bc_mp, dfops,
+				rcur->bc_private.a.agno,
+				agbno, aglen, XFS_RMAP_OWN_COW);
+		if (error)
+			return error;
+	}
+
+	return error;
+}
+
+/* Record a CoW staging extent in the refcount btree. */
+int
+xfs_refcount_alloc_cow_extent(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	xfs_fsblock_t			fsb,
+	xfs_extlen_t			len)
+{
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_ALLOC_COW,
+			fsb, len);
+}
+
+/* Forget a CoW staging event in the refcount btree. */
+int
+xfs_refcount_free_cow_extent(
+	struct xfs_mount		*mp,
+	struct xfs_defer_ops		*dfops,
+	xfs_fsblock_t			fsb,
+	xfs_extlen_t			len)
+{
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_FREE_COW,
+			fsb, len);
+}
+
+struct xfs_refcount_recovery {
+	struct list_head		rr_list;
+	struct xfs_refcount_irec	rr_rrec;
+};
+
+/* Stuff an extent on the recovery list. */
+STATIC int
+xfs_refcount_recover_extent(
+	struct xfs_btree_cur		*cur,
+	union xfs_btree_rec		*rec,
+	void				*priv)
+{
+	struct list_head		*debris = priv;
+	struct xfs_refcount_recovery	*rr;
+
+	if (be32_to_cpu(rec->refc.rc_refcount) != 1)
+		return -EFSCORRUPTED;
+
+	rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), KM_SLEEP);
+	xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
+	list_add_tail(&rr->rr_list, debris);
+
+	return 0;
+}
+
+/* Find and remove leftover CoW reservations. */
+int
+xfs_refcount_recover_cow_leftovers(
+	struct xfs_mount		*mp,
+	xfs_agnumber_t			agno)
+{
+	struct xfs_trans		*tp;
+	struct xfs_btree_cur		*cur;
+	struct xfs_buf			*agbp;
+	struct xfs_refcount_recovery	*rr, *n;
+	struct list_head		debris;
+	union xfs_btree_irec		low;
+	union xfs_btree_irec		high;
+	struct xfs_defer_ops		dfops;
+	xfs_fsblock_t			fsb;
+	xfs_agblock_t			agbno;
+	int				error;
+
+	if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
+		return -EOPNOTSUPP;
+
+	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
+	if (error)
+		return error;
+	cur = xfs_refcountbt_init_cursor(mp, NULL, agbp, agno, NULL);
+
+	/* Find all the leftover CoW staging extents. */
+	INIT_LIST_HEAD(&debris);
+	memset(&low, 0, sizeof(low));
+	memset(&high, 0, sizeof(high));
+	low.rc.rc_startblock = XFS_REFC_COW_START;
+	high.rc.rc_startblock = -1U;
+	error = xfs_btree_query_range(cur, &low, &high,
+			xfs_refcount_recover_extent, &debris);
+	if (error)
+		goto out_cursor;
+	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+	xfs_buf_relse(agbp);
+
+	/* Now iterate the list to free the leftovers */
+	list_for_each_entry(rr, &debris, rr_list) {
+		/* Set up transaction. */
+		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
+		if (error)
+			goto out_free;
+
+		trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
+
+		/* Free the orphan record */
+		xfs_defer_init(&dfops, &fsb);
+		agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
+		fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
+		error = xfs_refcount_free_cow_extent(mp, &dfops, fsb,
+				rr->rr_rrec.rc_blockcount);
+		if (error)
+			goto out_defer;
+
+		/* Free the block. */
+		xfs_bmap_add_free(mp, &dfops, fsb,
+				rr->rr_rrec.rc_blockcount, NULL);
+
+		error = xfs_defer_finish(&tp, &dfops, NULL);
+		if (error)
+			goto out_defer;
+
+		error = xfs_trans_commit(tp);
+		if (error)
+			goto out_free;
+	}
+
+out_free:
+	/* Free the leftover list */
+	list_for_each_entry_safe(rr, n, &debris, rr_list) {
+		list_del(&rr->rr_list);
+		kmem_free(rr);
+	}
+	return error;
+
+out_cursor:
+	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
+	xfs_buf_relse(agbp);
+	goto out_free;
+
+out_defer:
+	xfs_defer_cancel(&dfops);
+	xfs_trans_cancel(tp);
+	goto out_free;
+}

fs/xfs/libxfs/xfs_refcount.h

@@ -0,0 +1,70 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_REFCOUNT_H__
+#define __XFS_REFCOUNT_H__
+
+extern int xfs_refcount_lookup_le(struct xfs_btree_cur *cur,
+		xfs_agblock_t bno, int *stat);
+extern int xfs_refcount_lookup_ge(struct xfs_btree_cur *cur,
+		xfs_agblock_t bno, int *stat);
+extern int xfs_refcount_get_rec(struct xfs_btree_cur *cur,
+		struct xfs_refcount_irec *irec, int *stat);
+
+enum xfs_refcount_intent_type {
+	XFS_REFCOUNT_INCREASE = 1,
+	XFS_REFCOUNT_DECREASE,
+	XFS_REFCOUNT_ALLOC_COW,
+	XFS_REFCOUNT_FREE_COW,
+};
+
+struct xfs_refcount_intent {
+	struct list_head			ri_list;
+	enum xfs_refcount_intent_type		ri_type;
+	xfs_fsblock_t				ri_startblock;
+	xfs_extlen_t				ri_blockcount;
+};
+
+extern int xfs_refcount_increase_extent(struct xfs_mount *mp,
+		struct xfs_defer_ops *dfops, struct xfs_bmbt_irec *irec);
+extern int xfs_refcount_decrease_extent(struct xfs_mount *mp,
+		struct xfs_defer_ops *dfops, struct xfs_bmbt_irec *irec);
+
+extern void xfs_refcount_finish_one_cleanup(struct xfs_trans *tp,
+		struct xfs_btree_cur *rcur, int error);
+extern int xfs_refcount_finish_one(struct xfs_trans *tp,
+		struct xfs_defer_ops *dfops, enum xfs_refcount_intent_type type,
+		xfs_fsblock_t startblock, xfs_extlen_t blockcount,
+		xfs_fsblock_t *new_fsb, xfs_extlen_t *new_len,
+		struct xfs_btree_cur **pcur);
+
+extern int xfs_refcount_find_shared(struct xfs_btree_cur *cur,
+		xfs_agblock_t agbno, xfs_extlen_t aglen, xfs_agblock_t *fbno,
+		xfs_extlen_t *flen, bool find_end_of_shared);
+
+extern int xfs_refcount_alloc_cow_extent(struct xfs_mount *mp,
+		struct xfs_defer_ops *dfops, xfs_fsblock_t fsb,
+		xfs_extlen_t len);
+extern int xfs_refcount_free_cow_extent(struct xfs_mount *mp,
+		struct xfs_defer_ops *dfops, xfs_fsblock_t fsb,
+		xfs_extlen_t len);
+extern int xfs_refcount_recover_cow_leftovers(struct xfs_mount *mp,
+		xfs_agnumber_t agno);
+
+#endif	/* __XFS_REFCOUNT_H__ */

fs/xfs/libxfs/xfs_refcount_btree.c

@@ -0,0 +1,451 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_sb.h"
+#include "xfs_mount.h"
+#include "xfs_btree.h"
+#include "xfs_bmap.h"
+#include "xfs_refcount_btree.h"
+#include "xfs_alloc.h"
+#include "xfs_error.h"
+#include "xfs_trace.h"
+#include "xfs_cksum.h"
+#include "xfs_trans.h"
+#include "xfs_bit.h"
+#include "xfs_rmap.h"
+
+static struct xfs_btree_cur *
+xfs_refcountbt_dup_cursor(
+	struct xfs_btree_cur	*cur)
+{
+	return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
+			cur->bc_private.a.agbp, cur->bc_private.a.agno,
+			cur->bc_private.a.dfops);
+}
+
+STATIC void
+xfs_refcountbt_set_root(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*ptr,
+	int			inc)
+{
+	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
+	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
+	xfs_agnumber_t		seqno = be32_to_cpu(agf->agf_seqno);
+	struct xfs_perag	*pag = xfs_perag_get(cur->bc_mp, seqno);
+
+	ASSERT(ptr->s != 0);
+
+	agf->agf_refcount_root = ptr->s;
+	be32_add_cpu(&agf->agf_refcount_level, inc);
+	pag->pagf_refcount_level += inc;
+	xfs_perag_put(pag);
+
+	xfs_alloc_log_agf(cur->bc_tp, agbp,
+			XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
+}
+
+STATIC int
+xfs_refcountbt_alloc_block(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*start,
+	union xfs_btree_ptr	*new,
+	int			*stat)
+{
+	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
+	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
+	struct xfs_alloc_arg	args;		/* block allocation args */
+	int			error;		/* error return value */
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+
+	memset(&args, 0, sizeof(args));
+	args.tp = cur->bc_tp;
+	args.mp = cur->bc_mp;
+	args.type = XFS_ALLOCTYPE_NEAR_BNO;
+	args.fsbno = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_private.a.agno,
+			xfs_refc_block(args.mp));
+	args.firstblock = args.fsbno;
+	xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_REFC);
+	args.minlen = args.maxlen = args.prod = 1;
+	args.resv = XFS_AG_RESV_METADATA;
+
+	error = xfs_alloc_vextent(&args);
+	if (error)
+		goto out_error;
+	trace_xfs_refcountbt_alloc_block(cur->bc_mp, cur->bc_private.a.agno,
+			args.agbno, 1);
+	if (args.fsbno == NULLFSBLOCK) {
+		XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+		*stat = 0;
+		return 0;
+	}
+	ASSERT(args.agno == cur->bc_private.a.agno);
+	ASSERT(args.len == 1);
+
+	new->s = cpu_to_be32(args.agbno);
+	be32_add_cpu(&agf->agf_refcount_blocks, 1);
+	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+	*stat = 1;
+	return 0;
+
+out_error:
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+	return error;
+}
+
+STATIC int
+xfs_refcountbt_free_block(
+	struct xfs_btree_cur	*cur,
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
+	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
+	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
+	struct xfs_owner_info	oinfo;
+	int			error;
+
+	trace_xfs_refcountbt_free_block(cur->bc_mp, cur->bc_private.a.agno,
+			XFS_FSB_TO_AGBNO(cur->bc_mp, fsbno), 1);
+	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_REFC);
+	be32_add_cpu(&agf->agf_refcount_blocks, -1);
+	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
+	error = xfs_free_extent(cur->bc_tp, fsbno, 1, &oinfo,
+			XFS_AG_RESV_METADATA);
+	if (error)
+		return error;
+
+	return error;
+}
+
+STATIC int
+xfs_refcountbt_get_minrecs(
+	struct xfs_btree_cur	*cur,
+	int			level)
+{
+	return cur->bc_mp->m_refc_mnr[level != 0];
+}
+
+STATIC int
+xfs_refcountbt_get_maxrecs(
+	struct xfs_btree_cur	*cur,
+	int			level)
+{
+	return cur->bc_mp->m_refc_mxr[level != 0];
+}
+
+STATIC void
+xfs_refcountbt_init_key_from_rec(
+	union xfs_btree_key	*key,
+	union xfs_btree_rec	*rec)
+{
+	key->refc.rc_startblock = rec->refc.rc_startblock;
+}
+
+STATIC void
+xfs_refcountbt_init_high_key_from_rec(
+	union xfs_btree_key	*key,
+	union xfs_btree_rec	*rec)
+{
+	__u32			x;
+
+	x = be32_to_cpu(rec->refc.rc_startblock);
+	x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
+	key->refc.rc_startblock = cpu_to_be32(x);
+}
+
+STATIC void
+xfs_refcountbt_init_rec_from_cur(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_rec	*rec)
+{
+	rec->refc.rc_startblock = cpu_to_be32(cur->bc_rec.rc.rc_startblock);
+	rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
+	rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
+}
+
+STATIC void
+xfs_refcountbt_init_ptr_from_cur(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*ptr)
+{
+	struct xfs_agf		*agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
+
+	ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
+	ASSERT(agf->agf_refcount_root != 0);
+
+	ptr->s = agf->agf_refcount_root;
+}
+
+STATIC __int64_t
+xfs_refcountbt_key_diff(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_key	*key)
+{
+	struct xfs_refcount_irec	*rec = &cur->bc_rec.rc;
+	struct xfs_refcount_key		*kp = &key->refc;
+
+	return (__int64_t)be32_to_cpu(kp->rc_startblock) - rec->rc_startblock;
+}
+
+STATIC __int64_t
+xfs_refcountbt_diff_two_keys(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_key	*k1,
+	union xfs_btree_key	*k2)
+{
+	return (__int64_t)be32_to_cpu(k1->refc.rc_startblock) -
+			  be32_to_cpu(k2->refc.rc_startblock);
+}
+
+STATIC bool
+xfs_refcountbt_verify(
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = bp->b_target->bt_mount;
+	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
+	struct xfs_perag	*pag = bp->b_pag;
+	unsigned int		level;
+
+	if (block->bb_magic != cpu_to_be32(XFS_REFC_CRC_MAGIC))
+		return false;
+
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return false;
+	if (!xfs_btree_sblock_v5hdr_verify(bp))
+		return false;
+
+	level = be16_to_cpu(block->bb_level);
+	if (pag && pag->pagf_init) {
+		if (level >= pag->pagf_refcount_level)
+			return false;
+	} else if (level >= mp->m_refc_maxlevels)
+		return false;
+
+	return xfs_btree_sblock_verify(bp, mp->m_refc_mxr[level != 0]);
+}
+
+STATIC void
+xfs_refcountbt_read_verify(
+	struct xfs_buf	*bp)
+{
+	if (!xfs_btree_sblock_verify_crc(bp))
+		xfs_buf_ioerror(bp, -EFSBADCRC);
+	else if (!xfs_refcountbt_verify(bp))
+		xfs_buf_ioerror(bp, -EFSCORRUPTED);
+
+	if (bp->b_error) {
+		trace_xfs_btree_corrupt(bp, _RET_IP_);
+		xfs_verifier_error(bp);
+	}
+}
+
+STATIC void
+xfs_refcountbt_write_verify(
+	struct xfs_buf	*bp)
+{
+	if (!xfs_refcountbt_verify(bp)) {
+		trace_xfs_btree_corrupt(bp, _RET_IP_);
+		xfs_buf_ioerror(bp, -EFSCORRUPTED);
+		xfs_verifier_error(bp);
+		return;
+	}
+	xfs_btree_sblock_calc_crc(bp);
+
+}
+
+const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
+	.name			= "xfs_refcountbt",
+	.verify_read		= xfs_refcountbt_read_verify,
+	.verify_write		= xfs_refcountbt_write_verify,
+};
+
+#if defined(DEBUG) || defined(XFS_WARN)
+STATIC int
+xfs_refcountbt_keys_inorder(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_key	*k1,
+	union xfs_btree_key	*k2)
+{
+	return be32_to_cpu(k1->refc.rc_startblock) <
+	       be32_to_cpu(k2->refc.rc_startblock);
+}
+
+STATIC int
+xfs_refcountbt_recs_inorder(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_rec	*r1,
+	union xfs_btree_rec	*r2)
+{
+	return  be32_to_cpu(r1->refc.rc_startblock) +
+		be32_to_cpu(r1->refc.rc_blockcount) <=
+		be32_to_cpu(r2->refc.rc_startblock);
+}
+#endif
+
+static const struct xfs_btree_ops xfs_refcountbt_ops = {
+	.rec_len		= sizeof(struct xfs_refcount_rec),
+	.key_len		= sizeof(struct xfs_refcount_key),
+
+	.dup_cursor		= xfs_refcountbt_dup_cursor,
+	.set_root		= xfs_refcountbt_set_root,
+	.alloc_block		= xfs_refcountbt_alloc_block,
+	.free_block		= xfs_refcountbt_free_block,
+	.get_minrecs		= xfs_refcountbt_get_minrecs,
+	.get_maxrecs		= xfs_refcountbt_get_maxrecs,
+	.init_key_from_rec	= xfs_refcountbt_init_key_from_rec,
+	.init_high_key_from_rec	= xfs_refcountbt_init_high_key_from_rec,
+	.init_rec_from_cur	= xfs_refcountbt_init_rec_from_cur,
+	.init_ptr_from_cur	= xfs_refcountbt_init_ptr_from_cur,
+	.key_diff		= xfs_refcountbt_key_diff,
+	.buf_ops		= &xfs_refcountbt_buf_ops,
+	.diff_two_keys		= xfs_refcountbt_diff_two_keys,
+#if defined(DEBUG) || defined(XFS_WARN)
+	.keys_inorder		= xfs_refcountbt_keys_inorder,
+	.recs_inorder		= xfs_refcountbt_recs_inorder,
+#endif
+};
+
+/*
+ * Allocate a new refcount btree cursor.
+ */
+struct xfs_btree_cur *
+xfs_refcountbt_init_cursor(
+	struct xfs_mount	*mp,
+	struct xfs_trans	*tp,
+	struct xfs_buf		*agbp,
+	xfs_agnumber_t		agno,
+	struct xfs_defer_ops	*dfops)
+{
+	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
+	struct xfs_btree_cur	*cur;
+
+	ASSERT(agno != NULLAGNUMBER);
+	ASSERT(agno < mp->m_sb.sb_agcount);
+	cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
+
+	cur->bc_tp = tp;
+	cur->bc_mp = mp;
+	cur->bc_btnum = XFS_BTNUM_REFC;
+	cur->bc_blocklog = mp->m_sb.sb_blocklog;
+	cur->bc_ops = &xfs_refcountbt_ops;
+
+	cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
+
+	cur->bc_private.a.agbp = agbp;
+	cur->bc_private.a.agno = agno;
+	cur->bc_private.a.dfops = dfops;
+	cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
+
+	cur->bc_private.a.priv.refc.nr_ops = 0;
+	cur->bc_private.a.priv.refc.shape_changes = 0;
+
+	return cur;
+}
+
+/*
+ * Calculate the number of records in a refcount btree block.
+ */
+int
+xfs_refcountbt_maxrecs(
+	struct xfs_mount	*mp,
+	int			blocklen,
+	bool			leaf)
+{
+	blocklen -= XFS_REFCOUNT_BLOCK_LEN;
+
+	if (leaf)
+		return blocklen / sizeof(struct xfs_refcount_rec);
+	return blocklen / (sizeof(struct xfs_refcount_key) +
+			   sizeof(xfs_refcount_ptr_t));
+}
+
+/* Compute the maximum height of a refcount btree. */
+void
+xfs_refcountbt_compute_maxlevels(
+	struct xfs_mount		*mp)
+{
+	mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(mp,
+			mp->m_refc_mnr, mp->m_sb.sb_agblocks);
+}
+
+/* Calculate the refcount btree size for some records. */
+xfs_extlen_t
+xfs_refcountbt_calc_size(
+	struct xfs_mount	*mp,
+	unsigned long long	len)
+{
+	return xfs_btree_calc_size(mp, mp->m_refc_mnr, len);
+}
+
+/*
+ * Calculate the maximum refcount btree size.
+ */
+xfs_extlen_t
+xfs_refcountbt_max_size(
+	struct xfs_mount	*mp)
+{
+	/* Bail out if we're uninitialized, which can happen in mkfs. */
+	if (mp->m_refc_mxr[0] == 0)
+		return 0;
+
+	return xfs_refcountbt_calc_size(mp, mp->m_sb.sb_agblocks);
+}
+
+/*
+ * Figure out how many blocks to reserve and how many are used by this btree.
+ */
+int
+xfs_refcountbt_calc_reserves(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_extlen_t		*ask,
+	xfs_extlen_t		*used)
+{
+	struct xfs_buf		*agbp;
+	struct xfs_agf		*agf;
+	xfs_extlen_t		tree_len;
+	int			error;
+
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	*ask += xfs_refcountbt_max_size(mp);
+
+	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
+	if (error)
+		return error;
+
+	agf = XFS_BUF_TO_AGF(agbp);
+	tree_len = be32_to_cpu(agf->agf_refcount_blocks);
+	xfs_buf_relse(agbp);
+
+	*used += tree_len;
+
+	return error;
+}

fs/xfs/libxfs/xfs_refcount_btree.h

@@ -0,0 +1,74 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_REFCOUNT_BTREE_H__
+#define	__XFS_REFCOUNT_BTREE_H__
+
+/*
+ * Reference Count Btree on-disk structures
+ */
+
+struct xfs_buf;
+struct xfs_btree_cur;
+struct xfs_mount;
+
+/*
+ * Btree block header size
+ */
+#define XFS_REFCOUNT_BLOCK_LEN	XFS_BTREE_SBLOCK_CRC_LEN
+
+/*
+ * Record, key, and pointer address macros for btree blocks.
+ *
+ * (note that some of these may appear unused, but they are used in userspace)
+ */
+#define XFS_REFCOUNT_REC_ADDR(block, index) \
+	((struct xfs_refcount_rec *) \
+		((char *)(block) + \
+		 XFS_REFCOUNT_BLOCK_LEN + \
+		 (((index) - 1) * sizeof(struct xfs_refcount_rec))))
+
+#define XFS_REFCOUNT_KEY_ADDR(block, index) \
+	((struct xfs_refcount_key *) \
+		((char *)(block) + \
+		 XFS_REFCOUNT_BLOCK_LEN + \
+		 ((index) - 1) * sizeof(struct xfs_refcount_key)))
+
+#define XFS_REFCOUNT_PTR_ADDR(block, index, maxrecs) \
+	((xfs_refcount_ptr_t *) \
+		((char *)(block) + \
+		 XFS_REFCOUNT_BLOCK_LEN + \
+		 (maxrecs) * sizeof(struct xfs_refcount_key) + \
+		 ((index) - 1) * sizeof(xfs_refcount_ptr_t)))
+
+extern struct xfs_btree_cur *xfs_refcountbt_init_cursor(struct xfs_mount *mp,
+		struct xfs_trans *tp, struct xfs_buf *agbp, xfs_agnumber_t agno,
+		struct xfs_defer_ops *dfops);
+extern int xfs_refcountbt_maxrecs(struct xfs_mount *mp, int blocklen,
+		bool leaf);
+extern void xfs_refcountbt_compute_maxlevels(struct xfs_mount *mp);
+
+extern xfs_extlen_t xfs_refcountbt_calc_size(struct xfs_mount *mp,
+		unsigned long long len);
+extern xfs_extlen_t xfs_refcountbt_max_size(struct xfs_mount *mp);
+
+extern int xfs_refcountbt_calc_reserves(struct xfs_mount *mp,
+		xfs_agnumber_t agno, xfs_extlen_t *ask, xfs_extlen_t *used);
+
+#endif	/* __XFS_REFCOUNT_BTREE_H__ */

fs/xfs/libxfs/xfs_rmap.c

@@ -148,6 +148,37 @@ done:
 	return error;
 }
 
+STATIC int
+xfs_rmap_delete(
+	struct xfs_btree_cur	*rcur,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		len,
+	uint64_t		owner,
+	uint64_t		offset,
+	unsigned int		flags)
+{
+	int			i;
+	int			error;
+
+	trace_xfs_rmap_delete(rcur->bc_mp, rcur->bc_private.a.agno, agbno,
+			len, owner, offset, flags);
+
+	error = xfs_rmap_lookup_eq(rcur, agbno, len, owner, offset, flags, &i);
+	if (error)
+		goto done;
+	XFS_WANT_CORRUPTED_GOTO(rcur->bc_mp, i == 1, done);
+
+	error = xfs_btree_delete(rcur, &i);
+	if (error)
+		goto done;
+	XFS_WANT_CORRUPTED_GOTO(rcur->bc_mp, i == 1, done);
+done:
+	if (error)
+		trace_xfs_rmap_delete_error(rcur->bc_mp,
+				rcur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
 static int
 xfs_rmap_btrec_to_irec(
 	union xfs_btree_rec	*rec,
@@ -180,6 +211,160 @@ xfs_rmap_get_rec(
 	return xfs_rmap_btrec_to_irec(rec, irec);
 }
 
+struct xfs_find_left_neighbor_info {
+	struct xfs_rmap_irec	high;
+	struct xfs_rmap_irec	*irec;
+	int			*stat;
+};
+
+/* For each rmap given, figure out if it matches the key we want. */
+STATIC int
+xfs_rmap_find_left_neighbor_helper(
+	struct xfs_btree_cur	*cur,
+	struct xfs_rmap_irec	*rec,
+	void			*priv)
+{
+	struct xfs_find_left_neighbor_info	*info = priv;
+
+	trace_xfs_rmap_find_left_neighbor_candidate(cur->bc_mp,
+			cur->bc_private.a.agno, rec->rm_startblock,
+			rec->rm_blockcount, rec->rm_owner, rec->rm_offset,
+			rec->rm_flags);
+
+	if (rec->rm_owner != info->high.rm_owner)
+		return XFS_BTREE_QUERY_RANGE_CONTINUE;
+	if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) &&
+	    !(rec->rm_flags & XFS_RMAP_BMBT_BLOCK) &&
+	    rec->rm_offset + rec->rm_blockcount - 1 != info->high.rm_offset)
+		return XFS_BTREE_QUERY_RANGE_CONTINUE;
+
+	*info->irec = *rec;
+	*info->stat = 1;
+	return XFS_BTREE_QUERY_RANGE_ABORT;
+}
+
+/*
+ * Find the record to the left of the given extent, being careful only to
+ * return a match with the same owner and adjacent physical and logical
+ * block ranges.
+ */
+int
+xfs_rmap_find_left_neighbor(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	uint64_t		owner,
+	uint64_t		offset,
+	unsigned int		flags,
+	struct xfs_rmap_irec	*irec,
+	int			*stat)
+{
+	struct xfs_find_left_neighbor_info	info;
+	int			error;
+
+	*stat = 0;
+	if (bno == 0)
+		return 0;
+	info.high.rm_startblock = bno - 1;
+	info.high.rm_owner = owner;
+	if (!XFS_RMAP_NON_INODE_OWNER(owner) &&
+	    !(flags & XFS_RMAP_BMBT_BLOCK)) {
+		if (offset == 0)
+			return 0;
+		info.high.rm_offset = offset - 1;
+	} else
+		info.high.rm_offset = 0;
+	info.high.rm_flags = flags;
+	info.high.rm_blockcount = 0;
+	info.irec = irec;
+	info.stat = stat;
+
+	trace_xfs_rmap_find_left_neighbor_query(cur->bc_mp,
+			cur->bc_private.a.agno, bno, 0, owner, offset, flags);
+
+	error = xfs_rmap_query_range(cur, &info.high, &info.high,
+			xfs_rmap_find_left_neighbor_helper, &info);
+	if (error == XFS_BTREE_QUERY_RANGE_ABORT)
+		error = 0;
+	if (*stat)
+		trace_xfs_rmap_find_left_neighbor_result(cur->bc_mp,
+				cur->bc_private.a.agno, irec->rm_startblock,
+				irec->rm_blockcount, irec->rm_owner,
+				irec->rm_offset, irec->rm_flags);
+	return error;
+}
+
+/* For each rmap given, figure out if it matches the key we want. */
+STATIC int
+xfs_rmap_lookup_le_range_helper(
+	struct xfs_btree_cur	*cur,
+	struct xfs_rmap_irec	*rec,
+	void			*priv)
+{
+	struct xfs_find_left_neighbor_info	*info = priv;
+
+	trace_xfs_rmap_lookup_le_range_candidate(cur->bc_mp,
+			cur->bc_private.a.agno, rec->rm_startblock,
+			rec->rm_blockcount, rec->rm_owner, rec->rm_offset,
+			rec->rm_flags);
+
+	if (rec->rm_owner != info->high.rm_owner)
+		return XFS_BTREE_QUERY_RANGE_CONTINUE;
+	if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) &&
+	    !(rec->rm_flags & XFS_RMAP_BMBT_BLOCK) &&
+	    (rec->rm_offset > info->high.rm_offset ||
+	     rec->rm_offset + rec->rm_blockcount <= info->high.rm_offset))
+		return XFS_BTREE_QUERY_RANGE_CONTINUE;
+
+	*info->irec = *rec;
+	*info->stat = 1;
+	return XFS_BTREE_QUERY_RANGE_ABORT;
+}
+
+/*
+ * Find the record to the left of the given extent, being careful only to
+ * return a match with the same owner and overlapping physical and logical
+ * block ranges.  This is the overlapping-interval version of
+ * xfs_rmap_lookup_le.
+ */
+int
+xfs_rmap_lookup_le_range(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	uint64_t		owner,
+	uint64_t		offset,
+	unsigned int		flags,
+	struct xfs_rmap_irec	*irec,
+	int			*stat)
+{
+	struct xfs_find_left_neighbor_info	info;
+	int			error;
+
+	info.high.rm_startblock = bno;
+	info.high.rm_owner = owner;
+	if (!XFS_RMAP_NON_INODE_OWNER(owner) && !(flags & XFS_RMAP_BMBT_BLOCK))
+		info.high.rm_offset = offset;
+	else
+		info.high.rm_offset = 0;
+	info.high.rm_flags = flags;
+	info.high.rm_blockcount = 0;
+	*stat = 0;
+	info.irec = irec;
+	info.stat = stat;
+
+	trace_xfs_rmap_lookup_le_range(cur->bc_mp,
+			cur->bc_private.a.agno, bno, 0, owner, offset, flags);
+	error = xfs_rmap_query_range(cur, &info.high, &info.high,
+			xfs_rmap_lookup_le_range_helper, &info);
+	if (error == XFS_BTREE_QUERY_RANGE_ABORT)
+		error = 0;
+	if (*stat)
+		trace_xfs_rmap_lookup_le_range_result(cur->bc_mp,
+				cur->bc_private.a.agno, irec->rm_startblock,
+				irec->rm_blockcount, irec->rm_owner,
+				irec->rm_offset, irec->rm_flags);
+	return error;
+}
+
 /*
  * Find the extent in the rmap btree and remove it.
  *
@@ -1093,123 +1278,816 @@ done:
 	return error;
 }
 
-#undef	NEW
-#undef	LEFT
-#undef	RIGHT
-#undef	PREV
-
-struct xfs_rmap_query_range_info {
-	xfs_rmap_query_range_fn	fn;
-	void				*priv;
-};
-
-/* Format btree record and pass to our callback. */
+/*
+ * Convert an unwritten extent to a real extent or vice versa.  If there is no
+ * possibility of overlapping extents, delegate to the simpler convert
+ * function.
+ */
 STATIC int
-xfs_rmap_query_range_helper(
+xfs_rmap_convert_shared(
 	struct xfs_btree_cur	*cur,
-	union xfs_btree_rec	*rec,
-	void			*priv)
+	xfs_agblock_t		bno,
+	xfs_extlen_t		len,
+	bool			unwritten,
+	struct xfs_owner_info	*oinfo)
 {
-	struct xfs_rmap_query_range_info	*query = priv;
-	struct xfs_rmap_irec			irec;
-	int					error;
+	struct xfs_mount	*mp = cur->bc_mp;
+	struct xfs_rmap_irec	r[4];	/* neighbor extent entries */
+					/* left is 0, right is 1, prev is 2 */
+					/* new is 3 */
+	uint64_t		owner;
+	uint64_t		offset;
+	uint64_t		new_endoff;
+	unsigned int		oldext;
+	unsigned int		newext;
+	unsigned int		flags = 0;
+	int			i;
+	int			state = 0;
+	int			error;
 
-	error = xfs_rmap_btrec_to_irec(rec, &irec);
-	if (error)
-		return error;
-	return query->fn(cur, &irec, query->priv);
-}
+	xfs_owner_info_unpack(oinfo, &owner, &offset, &flags);
+	ASSERT(!(XFS_RMAP_NON_INODE_OWNER(owner) ||
+			(flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK))));
+	oldext = unwritten ? XFS_RMAP_UNWRITTEN : 0;
+	new_endoff = offset + len;
+	trace_xfs_rmap_convert(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
 
-/* Find all rmaps between two keys. */
-int
-xfs_rmap_query_range(
-	struct xfs_btree_cur		*cur,
-	struct xfs_rmap_irec		*low_rec,
-	struct xfs_rmap_irec		*high_rec,
-	xfs_rmap_query_range_fn	fn,
-	void				*priv)
-{
-	union xfs_btree_irec		low_brec;
-	union xfs_btree_irec		high_brec;
-	struct xfs_rmap_query_range_info	query;
+	/*
+	 * For the initial lookup, look for and exact match or the left-adjacent
+	 * record for our insertion point. This will also give us the record for
+	 * start block contiguity tests.
+	 */
+	error = xfs_rmap_lookup_le_range(cur, bno, owner, offset, flags,
+			&PREV, &i);
+	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
 
-	low_brec.r = *low_rec;
-	high_brec.r = *high_rec;
-	query.priv = priv;
-	query.fn = fn;
-	return xfs_btree_query_range(cur, &low_brec, &high_brec,
-			xfs_rmap_query_range_helper, &query);
-}
+	ASSERT(PREV.rm_offset <= offset);
+	ASSERT(PREV.rm_offset + PREV.rm_blockcount >= new_endoff);
+	ASSERT((PREV.rm_flags & XFS_RMAP_UNWRITTEN) == oldext);
+	newext = ~oldext & XFS_RMAP_UNWRITTEN;
 
-/* Clean up after calling xfs_rmap_finish_one. */
-void
-xfs_rmap_finish_one_cleanup(
-	struct xfs_trans	*tp,
-	struct xfs_btree_cur	*rcur,
-	int			error)
-{
-	struct xfs_buf		*agbp;
+	/*
+	 * Set flags determining what part of the previous oldext allocation
+	 * extent is being replaced by a newext allocation.
+	 */
+	if (PREV.rm_offset == offset)
+		state |= RMAP_LEFT_FILLING;
+	if (PREV.rm_offset + PREV.rm_blockcount == new_endoff)
+		state |= RMAP_RIGHT_FILLING;
 
-	if (rcur == NULL)
-		return;
-	agbp = rcur->bc_private.a.agbp;
-	xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
+	/* Is there a left record that abuts our range? */
+	error = xfs_rmap_find_left_neighbor(cur, bno, owner, offset, newext,
+			&LEFT, &i);
 	if (error)
-		xfs_trans_brelse(tp, agbp);
-}
-
-/*
- * Process one of the deferred rmap operations.  We pass back the
- * btree cursor to maintain our lock on the rmapbt between calls.
- * This saves time and eliminates a buffer deadlock between the
- * superblock and the AGF because we'll always grab them in the same
- * order.
- */
-int
-xfs_rmap_finish_one(
-	struct xfs_trans		*tp,
-	enum xfs_rmap_intent_type	type,
-	__uint64_t			owner,
-	int				whichfork,
-	xfs_fileoff_t			startoff,
-	xfs_fsblock_t			startblock,
-	xfs_filblks_t			blockcount,
-	xfs_exntst_t			state,
-	struct xfs_btree_cur		**pcur)
-{
-	struct xfs_mount		*mp = tp->t_mountp;
-	struct xfs_btree_cur		*rcur;
-	struct xfs_buf			*agbp = NULL;
-	int				error = 0;
-	xfs_agnumber_t			agno;
-	struct xfs_owner_info		oinfo;
-	xfs_agblock_t			bno;
-	bool				unwritten;
-
-	agno = XFS_FSB_TO_AGNO(mp, startblock);
-	ASSERT(agno != NULLAGNUMBER);
-	bno = XFS_FSB_TO_AGBNO(mp, startblock);
+		goto done;
+	if (i) {
+		state |= RMAP_LEFT_VALID;
+		XFS_WANT_CORRUPTED_GOTO(mp,
+				LEFT.rm_startblock + LEFT.rm_blockcount <= bno,
+				done);
+		if (xfs_rmap_is_mergeable(&LEFT, owner, newext))
+			state |= RMAP_LEFT_CONTIG;
+	}
 
-	trace_xfs_rmap_deferred(mp, agno, type, bno, owner, whichfork,
-			startoff, blockcount, state);
+	/* Is there a right record that abuts our range? */
+	error = xfs_rmap_lookup_eq(cur, bno + len, len, owner, offset + len,
+			newext, &i);
+	if (error)
+		goto done;
+	if (i) {
+		state |= RMAP_RIGHT_VALID;
+		error = xfs_rmap_get_rec(cur, &RIGHT, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		XFS_WANT_CORRUPTED_GOTO(mp, bno + len <= RIGHT.rm_startblock,
+				done);
+		trace_xfs_rmap_find_right_neighbor_result(cur->bc_mp,
+				cur->bc_private.a.agno, RIGHT.rm_startblock,
+				RIGHT.rm_blockcount, RIGHT.rm_owner,
+				RIGHT.rm_offset, RIGHT.rm_flags);
+		if (xfs_rmap_is_mergeable(&RIGHT, owner, newext))
+			state |= RMAP_RIGHT_CONTIG;
+	}
 
-	if (XFS_TEST_ERROR(false, mp,
-			XFS_ERRTAG_RMAP_FINISH_ONE,
-			XFS_RANDOM_RMAP_FINISH_ONE))
-		return -EIO;
+	/* check that left + prev + right is not too long */
+	if ((state & (RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG |
+			 RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG)) ==
+	    (RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG |
+	     RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG) &&
+	    (unsigned long)LEFT.rm_blockcount + len +
+	     RIGHT.rm_blockcount > XFS_RMAP_LEN_MAX)
+		state &= ~RMAP_RIGHT_CONTIG;
 
+	trace_xfs_rmap_convert_state(mp, cur->bc_private.a.agno, state,
+			_RET_IP_);
 	/*
-	 * If we haven't gotten a cursor or the cursor AG doesn't match
-	 * the startblock, get one now.
+	 * Switch out based on the FILLING and CONTIG state bits.
 	 */
-	rcur = *pcur;
-	if (rcur != NULL && rcur->bc_private.a.agno != agno) {
-		xfs_rmap_finish_one_cleanup(tp, rcur, 0);
-		rcur = NULL;
-		*pcur = NULL;
-	}
-	if (rcur == NULL) {
-		/*
+	switch (state & (RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG |
+			 RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG)) {
+	case RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG |
+	     RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG:
+		/*
+		 * Setting all of a previous oldext extent to newext.
+		 * The left and right neighbors are both contiguous with new.
+		 */
+		error = xfs_rmap_delete(cur, RIGHT.rm_startblock,
+				RIGHT.rm_blockcount, RIGHT.rm_owner,
+				RIGHT.rm_offset, RIGHT.rm_flags);
+		if (error)
+			goto done;
+		error = xfs_rmap_delete(cur, PREV.rm_startblock,
+				PREV.rm_blockcount, PREV.rm_owner,
+				PREV.rm_offset, PREV.rm_flags);
+		if (error)
+			goto done;
+		NEW = LEFT;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount += PREV.rm_blockcount + RIGHT.rm_blockcount;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING | RMAP_RIGHT_FILLING | RMAP_LEFT_CONTIG:
+		/*
+		 * Setting all of a previous oldext extent to newext.
+		 * The left neighbor is contiguous, the right is not.
+		 */
+		error = xfs_rmap_delete(cur, PREV.rm_startblock,
+				PREV.rm_blockcount, PREV.rm_owner,
+				PREV.rm_offset, PREV.rm_flags);
+		if (error)
+			goto done;
+		NEW = LEFT;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount += PREV.rm_blockcount;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING | RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG:
+		/*
+		 * Setting all of a previous oldext extent to newext.
+		 * The right neighbor is contiguous, the left is not.
+		 */
+		error = xfs_rmap_delete(cur, RIGHT.rm_startblock,
+				RIGHT.rm_blockcount, RIGHT.rm_owner,
+				RIGHT.rm_offset, RIGHT.rm_flags);
+		if (error)
+			goto done;
+		NEW = PREV;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount += RIGHT.rm_blockcount;
+		NEW.rm_flags = RIGHT.rm_flags;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING | RMAP_RIGHT_FILLING:
+		/*
+		 * Setting all of a previous oldext extent to newext.
+		 * Neither the left nor right neighbors are contiguous with
+		 * the new one.
+		 */
+		NEW = PREV;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_flags = newext;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG:
+		/*
+		 * Setting the first part of a previous oldext extent to newext.
+		 * The left neighbor is contiguous.
+		 */
+		NEW = PREV;
+		error = xfs_rmap_delete(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		NEW.rm_offset += len;
+		NEW.rm_startblock += len;
+		NEW.rm_blockcount -= len;
+		error = xfs_rmap_insert(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		NEW = LEFT;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount += len;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING:
+		/*
+		 * Setting the first part of a previous oldext extent to newext.
+		 * The left neighbor is not contiguous.
+		 */
+		NEW = PREV;
+		error = xfs_rmap_delete(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		NEW.rm_offset += len;
+		NEW.rm_startblock += len;
+		NEW.rm_blockcount -= len;
+		error = xfs_rmap_insert(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		error = xfs_rmap_insert(cur, bno, len, owner, offset, newext);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_RIGHT_FILLING | RMAP_RIGHT_CONTIG:
+		/*
+		 * Setting the last part of a previous oldext extent to newext.
+		 * The right neighbor is contiguous with the new allocation.
+		 */
+		NEW = PREV;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount = offset - NEW.rm_offset;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		NEW = RIGHT;
+		error = xfs_rmap_delete(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		NEW.rm_offset = offset;
+		NEW.rm_startblock = bno;
+		NEW.rm_blockcount += len;
+		error = xfs_rmap_insert(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_RIGHT_FILLING:
+		/*
+		 * Setting the last part of a previous oldext extent to newext.
+		 * The right neighbor is not contiguous.
+		 */
+		NEW = PREV;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount -= len;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		error = xfs_rmap_insert(cur, bno, len, owner, offset, newext);
+		if (error)
+			goto done;
+		break;
+
+	case 0:
+		/*
+		 * Setting the middle part of a previous oldext extent to
+		 * newext.  Contiguity is impossible here.
+		 * One extent becomes three extents.
+		 */
+		/* new right extent - oldext */
+		NEW.rm_startblock = bno + len;
+		NEW.rm_owner = owner;
+		NEW.rm_offset = new_endoff;
+		NEW.rm_blockcount = PREV.rm_offset + PREV.rm_blockcount -
+				new_endoff;
+		NEW.rm_flags = PREV.rm_flags;
+		error = xfs_rmap_insert(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner, NEW.rm_offset,
+				NEW.rm_flags);
+		if (error)
+			goto done;
+		/* new left extent - oldext */
+		NEW = PREV;
+		error = xfs_rmap_lookup_eq(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner,
+				NEW.rm_offset, NEW.rm_flags, &i);
+		if (error)
+			goto done;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
+		NEW.rm_blockcount = offset - NEW.rm_offset;
+		error = xfs_rmap_update(cur, &NEW);
+		if (error)
+			goto done;
+		/* new middle extent - newext */
+		NEW.rm_startblock = bno;
+		NEW.rm_blockcount = len;
+		NEW.rm_owner = owner;
+		NEW.rm_offset = offset;
+		NEW.rm_flags = newext;
+		error = xfs_rmap_insert(cur, NEW.rm_startblock,
+				NEW.rm_blockcount, NEW.rm_owner, NEW.rm_offset,
+				NEW.rm_flags);
+		if (error)
+			goto done;
+		break;
+
+	case RMAP_LEFT_FILLING | RMAP_LEFT_CONTIG | RMAP_RIGHT_CONTIG:
+	case RMAP_RIGHT_FILLING | RMAP_LEFT_CONTIG | RMAP_RIGHT_CONTIG:
+	case RMAP_LEFT_FILLING | RMAP_RIGHT_CONTIG:
+	case RMAP_RIGHT_FILLING | RMAP_LEFT_CONTIG:
+	case RMAP_LEFT_CONTIG | RMAP_RIGHT_CONTIG:
+	case RMAP_LEFT_CONTIG:
+	case RMAP_RIGHT_CONTIG:
+		/*
+		 * These cases are all impossible.
+		 */
+		ASSERT(0);
+	}
+
+	trace_xfs_rmap_convert_done(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
+done:
+	if (error)
+		trace_xfs_rmap_convert_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+#undef	NEW
+#undef	LEFT
+#undef	RIGHT
+#undef	PREV
+
+/*
+ * Find an extent in the rmap btree and unmap it.  For rmap extent types that
+ * can overlap (data fork rmaps on reflink filesystems) we must be careful
+ * that the prev/next records in the btree might belong to another owner.
+ * Therefore we must use delete+insert to alter any of the key fields.
+ *
+ * For every other situation there can only be one owner for a given extent,
+ * so we can call the regular _free function.
+ */
+STATIC int
+xfs_rmap_unmap_shared(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	xfs_extlen_t		len,
+	bool			unwritten,
+	struct xfs_owner_info	*oinfo)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	struct xfs_rmap_irec	ltrec;
+	uint64_t		ltoff;
+	int			error = 0;
+	int			i;
+	uint64_t		owner;
+	uint64_t		offset;
+	unsigned int		flags;
+
+	xfs_owner_info_unpack(oinfo, &owner, &offset, &flags);
+	if (unwritten)
+		flags |= XFS_RMAP_UNWRITTEN;
+	trace_xfs_rmap_unmap(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
+
+	/*
+	 * We should always have a left record because there's a static record
+	 * for the AG headers at rm_startblock == 0 created by mkfs/growfs that
+	 * will not ever be removed from the tree.
+	 */
+	error = xfs_rmap_lookup_le_range(cur, bno, owner, offset, flags,
+			&ltrec, &i);
+	if (error)
+		goto out_error;
+	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_error);
+	ltoff = ltrec.rm_offset;
+
+	/* Make sure the extent we found covers the entire freeing range. */
+	XFS_WANT_CORRUPTED_GOTO(mp, ltrec.rm_startblock <= bno &&
+		ltrec.rm_startblock + ltrec.rm_blockcount >=
+		bno + len, out_error);
+
+	/* Make sure the owner matches what we expect to find in the tree. */
+	XFS_WANT_CORRUPTED_GOTO(mp, owner == ltrec.rm_owner, out_error);
+
+	/* Make sure the unwritten flag matches. */
+	XFS_WANT_CORRUPTED_GOTO(mp, (flags & XFS_RMAP_UNWRITTEN) ==
+			(ltrec.rm_flags & XFS_RMAP_UNWRITTEN), out_error);
+
+	/* Check the offset. */
+	XFS_WANT_CORRUPTED_GOTO(mp, ltrec.rm_offset <= offset, out_error);
+	XFS_WANT_CORRUPTED_GOTO(mp, offset <= ltoff + ltrec.rm_blockcount,
+			out_error);
+
+	if (ltrec.rm_startblock == bno && ltrec.rm_blockcount == len) {
+		/* Exact match, simply remove the record from rmap tree. */
+		error = xfs_rmap_delete(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags);
+		if (error)
+			goto out_error;
+	} else if (ltrec.rm_startblock == bno) {
+		/*
+		 * Overlap left hand side of extent: move the start, trim the
+		 * length and update the current record.
+		 *
+		 *       ltbno                ltlen
+		 * Orig:    |oooooooooooooooooooo|
+		 * Freeing: |fffffffff|
+		 * Result:            |rrrrrrrrrr|
+		 *         bno       len
+		 */
+
+		/* Delete prev rmap. */
+		error = xfs_rmap_delete(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags);
+		if (error)
+			goto out_error;
+
+		/* Add an rmap at the new offset. */
+		ltrec.rm_startblock += len;
+		ltrec.rm_blockcount -= len;
+		ltrec.rm_offset += len;
+		error = xfs_rmap_insert(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags);
+		if (error)
+			goto out_error;
+	} else if (ltrec.rm_startblock + ltrec.rm_blockcount == bno + len) {
+		/*
+		 * Overlap right hand side of extent: trim the length and
+		 * update the current record.
+		 *
+		 *       ltbno                ltlen
+		 * Orig:    |oooooooooooooooooooo|
+		 * Freeing:            |fffffffff|
+		 * Result:  |rrrrrrrrrr|
+		 *                    bno       len
+		 */
+		error = xfs_rmap_lookup_eq(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags, &i);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_error);
+		ltrec.rm_blockcount -= len;
+		error = xfs_rmap_update(cur, &ltrec);
+		if (error)
+			goto out_error;
+	} else {
+		/*
+		 * Overlap middle of extent: trim the length of the existing
+		 * record to the length of the new left-extent size, increment
+		 * the insertion position so we can insert a new record
+		 * containing the remaining right-extent space.
+		 *
+		 *       ltbno                ltlen
+		 * Orig:    |oooooooooooooooooooo|
+		 * Freeing:       |fffffffff|
+		 * Result:  |rrrrr|         |rrrr|
+		 *               bno       len
+		 */
+		xfs_extlen_t	orig_len = ltrec.rm_blockcount;
+
+		/* Shrink the left side of the rmap */
+		error = xfs_rmap_lookup_eq(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags, &i);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_error);
+		ltrec.rm_blockcount = bno - ltrec.rm_startblock;
+		error = xfs_rmap_update(cur, &ltrec);
+		if (error)
+			goto out_error;
+
+		/* Add an rmap at the new offset */
+		error = xfs_rmap_insert(cur, bno + len,
+				orig_len - len - ltrec.rm_blockcount,
+				ltrec.rm_owner, offset + len,
+				ltrec.rm_flags);
+		if (error)
+			goto out_error;
+	}
+
+	trace_xfs_rmap_unmap_done(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
+out_error:
+	if (error)
+		trace_xfs_rmap_unmap_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Find an extent in the rmap btree and map it.  For rmap extent types that
+ * can overlap (data fork rmaps on reflink filesystems) we must be careful
+ * that the prev/next records in the btree might belong to another owner.
+ * Therefore we must use delete+insert to alter any of the key fields.
+ *
+ * For every other situation there can only be one owner for a given extent,
+ * so we can call the regular _alloc function.
+ */
+STATIC int
+xfs_rmap_map_shared(
+	struct xfs_btree_cur	*cur,
+	xfs_agblock_t		bno,
+	xfs_extlen_t		len,
+	bool			unwritten,
+	struct xfs_owner_info	*oinfo)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	struct xfs_rmap_irec	ltrec;
+	struct xfs_rmap_irec	gtrec;
+	int			have_gt;
+	int			have_lt;
+	int			error = 0;
+	int			i;
+	uint64_t		owner;
+	uint64_t		offset;
+	unsigned int		flags = 0;
+
+	xfs_owner_info_unpack(oinfo, &owner, &offset, &flags);
+	if (unwritten)
+		flags |= XFS_RMAP_UNWRITTEN;
+	trace_xfs_rmap_map(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
+
+	/* Is there a left record that abuts our range? */
+	error = xfs_rmap_find_left_neighbor(cur, bno, owner, offset, flags,
+			&ltrec, &have_lt);
+	if (error)
+		goto out_error;
+	if (have_lt &&
+	    !xfs_rmap_is_mergeable(&ltrec, owner, flags))
+		have_lt = 0;
+
+	/* Is there a right record that abuts our range? */
+	error = xfs_rmap_lookup_eq(cur, bno + len, len, owner, offset + len,
+			flags, &have_gt);
+	if (error)
+		goto out_error;
+	if (have_gt) {
+		error = xfs_rmap_get_rec(cur, &gtrec, &have_gt);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(mp, have_gt == 1, out_error);
+		trace_xfs_rmap_find_right_neighbor_result(cur->bc_mp,
+			cur->bc_private.a.agno, gtrec.rm_startblock,
+			gtrec.rm_blockcount, gtrec.rm_owner,
+			gtrec.rm_offset, gtrec.rm_flags);
+
+		if (!xfs_rmap_is_mergeable(&gtrec, owner, flags))
+			have_gt = 0;
+	}
+
+	if (have_lt &&
+	    ltrec.rm_startblock + ltrec.rm_blockcount == bno &&
+	    ltrec.rm_offset + ltrec.rm_blockcount == offset) {
+		/*
+		 * Left edge contiguous, merge into left record.
+		 *
+		 *       ltbno     ltlen
+		 * orig:   |ooooooooo|
+		 * adding:           |aaaaaaaaa|
+		 * result: |rrrrrrrrrrrrrrrrrrr|
+		 *                  bno       len
+		 */
+		ltrec.rm_blockcount += len;
+		if (have_gt &&
+		    bno + len == gtrec.rm_startblock &&
+		    offset + len == gtrec.rm_offset) {
+			/*
+			 * Right edge also contiguous, delete right record
+			 * and merge into left record.
+			 *
+			 *       ltbno     ltlen    gtbno     gtlen
+			 * orig:   |ooooooooo|         |ooooooooo|
+			 * adding:           |aaaaaaaaa|
+			 * result: |rrrrrrrrrrrrrrrrrrrrrrrrrrrrr|
+			 */
+			ltrec.rm_blockcount += gtrec.rm_blockcount;
+			error = xfs_rmap_delete(cur, gtrec.rm_startblock,
+					gtrec.rm_blockcount, gtrec.rm_owner,
+					gtrec.rm_offset, gtrec.rm_flags);
+			if (error)
+				goto out_error;
+		}
+
+		/* Point the cursor back to the left record and update. */
+		error = xfs_rmap_lookup_eq(cur, ltrec.rm_startblock,
+				ltrec.rm_blockcount, ltrec.rm_owner,
+				ltrec.rm_offset, ltrec.rm_flags, &i);
+		if (error)
+			goto out_error;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_error);
+
+		error = xfs_rmap_update(cur, &ltrec);
+		if (error)
+			goto out_error;
+	} else if (have_gt &&
+		   bno + len == gtrec.rm_startblock &&
+		   offset + len == gtrec.rm_offset) {
+		/*
+		 * Right edge contiguous, merge into right record.
+		 *
+		 *                 gtbno     gtlen
+		 * Orig:             |ooooooooo|
+		 * adding: |aaaaaaaaa|
+		 * Result: |rrrrrrrrrrrrrrrrrrr|
+		 *        bno       len
+		 */
+		/* Delete the old record. */
+		error = xfs_rmap_delete(cur, gtrec.rm_startblock,
+				gtrec.rm_blockcount, gtrec.rm_owner,
+				gtrec.rm_offset, gtrec.rm_flags);
+		if (error)
+			goto out_error;
+
+		/* Move the start and re-add it. */
+		gtrec.rm_startblock = bno;
+		gtrec.rm_blockcount += len;
+		gtrec.rm_offset = offset;
+		error = xfs_rmap_insert(cur, gtrec.rm_startblock,
+				gtrec.rm_blockcount, gtrec.rm_owner,
+				gtrec.rm_offset, gtrec.rm_flags);
+		if (error)
+			goto out_error;
+	} else {
+		/*
+		 * No contiguous edge with identical owner, insert
+		 * new record at current cursor position.
+		 */
+		error = xfs_rmap_insert(cur, bno, len, owner, offset, flags);
+		if (error)
+			goto out_error;
+	}
+
+	trace_xfs_rmap_map_done(mp, cur->bc_private.a.agno, bno, len,
+			unwritten, oinfo);
+out_error:
+	if (error)
+		trace_xfs_rmap_map_error(cur->bc_mp,
+				cur->bc_private.a.agno, error, _RET_IP_);
+	return error;
+}
+
+struct xfs_rmap_query_range_info {
+	xfs_rmap_query_range_fn	fn;
+	void				*priv;
+};
+
+/* Format btree record and pass to our callback. */
+STATIC int
+xfs_rmap_query_range_helper(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_rec	*rec,
+	void			*priv)
+{
+	struct xfs_rmap_query_range_info	*query = priv;
+	struct xfs_rmap_irec			irec;
+	int					error;
+
+	error = xfs_rmap_btrec_to_irec(rec, &irec);
+	if (error)
+		return error;
+	return query->fn(cur, &irec, query->priv);
+}
+
+/* Find all rmaps between two keys. */
+int
+xfs_rmap_query_range(
+	struct xfs_btree_cur		*cur,
+	struct xfs_rmap_irec		*low_rec,
+	struct xfs_rmap_irec		*high_rec,
+	xfs_rmap_query_range_fn	fn,
+	void				*priv)
+{
+	union xfs_btree_irec		low_brec;
+	union xfs_btree_irec		high_brec;
+	struct xfs_rmap_query_range_info	query;
+
+	low_brec.r = *low_rec;
+	high_brec.r = *high_rec;
+	query.priv = priv;
+	query.fn = fn;
+	return xfs_btree_query_range(cur, &low_brec, &high_brec,
+			xfs_rmap_query_range_helper, &query);
+}
+
+/* Clean up after calling xfs_rmap_finish_one. */
+void
+xfs_rmap_finish_one_cleanup(
+	struct xfs_trans	*tp,
+	struct xfs_btree_cur	*rcur,
+	int			error)
+{
+	struct xfs_buf		*agbp;
+
+	if (rcur == NULL)
+		return;
+	agbp = rcur->bc_private.a.agbp;
+	xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
+	if (error)
+		xfs_trans_brelse(tp, agbp);
+}
+
+/*
+ * Process one of the deferred rmap operations.  We pass back the
+ * btree cursor to maintain our lock on the rmapbt between calls.
+ * This saves time and eliminates a buffer deadlock between the
+ * superblock and the AGF because we'll always grab them in the same
+ * order.
+ */
+int
+xfs_rmap_finish_one(
+	struct xfs_trans		*tp,
+	enum xfs_rmap_intent_type	type,
+	__uint64_t			owner,
+	int				whichfork,
+	xfs_fileoff_t			startoff,
+	xfs_fsblock_t			startblock,
+	xfs_filblks_t			blockcount,
+	xfs_exntst_t			state,
+	struct xfs_btree_cur		**pcur)
+{
+	struct xfs_mount		*mp = tp->t_mountp;
+	struct xfs_btree_cur		*rcur;
+	struct xfs_buf			*agbp = NULL;
+	int				error = 0;
+	xfs_agnumber_t			agno;
+	struct xfs_owner_info		oinfo;
+	xfs_agblock_t			bno;
+	bool				unwritten;
+
+	agno = XFS_FSB_TO_AGNO(mp, startblock);
+	ASSERT(agno != NULLAGNUMBER);
+	bno = XFS_FSB_TO_AGBNO(mp, startblock);
+
+	trace_xfs_rmap_deferred(mp, agno, type, bno, owner, whichfork,
+			startoff, blockcount, state);
+
+	if (XFS_TEST_ERROR(false, mp,
+			XFS_ERRTAG_RMAP_FINISH_ONE,
+			XFS_RANDOM_RMAP_FINISH_ONE))
+		return -EIO;
+
+	/*
+	 * If we haven't gotten a cursor or the cursor AG doesn't match
+	 * the startblock, get one now.
+	 */
+	rcur = *pcur;
+	if (rcur != NULL && rcur->bc_private.a.agno != agno) {
+		xfs_rmap_finish_one_cleanup(tp, rcur, 0);
+		rcur = NULL;
+		*pcur = NULL;
+	}
+	if (rcur == NULL) {
+		/*
 		 * Refresh the freelist before we start changing the
 		 * rmapbt, because a shape change could cause us to
 		 * allocate blocks.
@@ -1237,15 +2115,27 @@ xfs_rmap_finish_one(
 	case XFS_RMAP_MAP:
 		error = xfs_rmap_map(rcur, bno, blockcount, unwritten, &oinfo);
 		break;
+	case XFS_RMAP_MAP_SHARED:
+		error = xfs_rmap_map_shared(rcur, bno, blockcount, unwritten,
+				&oinfo);
+		break;
 	case XFS_RMAP_FREE:
 	case XFS_RMAP_UNMAP:
 		error = xfs_rmap_unmap(rcur, bno, blockcount, unwritten,
 				&oinfo);
 		break;
+	case XFS_RMAP_UNMAP_SHARED:
+		error = xfs_rmap_unmap_shared(rcur, bno, blockcount, unwritten,
+				&oinfo);
+		break;
 	case XFS_RMAP_CONVERT:
 		error = xfs_rmap_convert(rcur, bno, blockcount, !unwritten,
 				&oinfo);
 		break;
+	case XFS_RMAP_CONVERT_SHARED:
+		error = xfs_rmap_convert_shared(rcur, bno, blockcount,
+				!unwritten, &oinfo);
+		break;
 	default:
 		ASSERT(0);
 		error = -EFSCORRUPTED;
@@ -1263,9 +2153,10 @@ out_cur:
  */
 static bool
 xfs_rmap_update_is_needed(
-	struct xfs_mount	*mp)
+	struct xfs_mount	*mp,
+	int			whichfork)
 {
-	return xfs_sb_version_hasrmapbt(&mp->m_sb);
+	return xfs_sb_version_hasrmapbt(&mp->m_sb) && whichfork != XFS_COW_FORK;
 }
 
 /*
@@ -1311,10 +2202,11 @@ xfs_rmap_map_extent(
 	int			whichfork,
 	struct xfs_bmbt_irec	*PREV)
 {
-	if (!xfs_rmap_update_is_needed(mp))
+	if (!xfs_rmap_update_is_needed(mp, whichfork))
 		return 0;
 
-	return __xfs_rmap_add(mp, dfops, XFS_RMAP_MAP, ip->i_ino,
+	return __xfs_rmap_add(mp, dfops, xfs_is_reflink_inode(ip) ?
+			XFS_RMAP_MAP_SHARED : XFS_RMAP_MAP, ip->i_ino,
 			whichfork, PREV);
 }
 
@@ -1327,10 +2219,11 @@ xfs_rmap_unmap_extent(
 	int			whichfork,
 	struct xfs_bmbt_irec	*PREV)
 {
-	if (!xfs_rmap_update_is_needed(mp))
+	if (!xfs_rmap_update_is_needed(mp, whichfork))
 		return 0;
 
-	return __xfs_rmap_add(mp, dfops, XFS_RMAP_UNMAP, ip->i_ino,
+	return __xfs_rmap_add(mp, dfops, xfs_is_reflink_inode(ip) ?
+			XFS_RMAP_UNMAP_SHARED : XFS_RMAP_UNMAP, ip->i_ino,
 			whichfork, PREV);
 }
 
@@ -1343,10 +2236,11 @@ xfs_rmap_convert_extent(
 	int			whichfork,
 	struct xfs_bmbt_irec	*PREV)
 {
-	if (!xfs_rmap_update_is_needed(mp))
+	if (!xfs_rmap_update_is_needed(mp, whichfork))
 		return 0;
 
-	return __xfs_rmap_add(mp, dfops, XFS_RMAP_CONVERT, ip->i_ino,
+	return __xfs_rmap_add(mp, dfops, xfs_is_reflink_inode(ip) ?
+			XFS_RMAP_CONVERT_SHARED : XFS_RMAP_CONVERT, ip->i_ino,
 			whichfork, PREV);
 }
 
@@ -1362,7 +2256,7 @@ xfs_rmap_alloc_extent(
 {
 	struct xfs_bmbt_irec	bmap;
 
-	if (!xfs_rmap_update_is_needed(mp))
+	if (!xfs_rmap_update_is_needed(mp, XFS_DATA_FORK))
 		return 0;
 
 	bmap.br_startblock = XFS_AGB_TO_FSB(mp, agno, bno);
@@ -1386,7 +2280,7 @@ xfs_rmap_free_extent(
 {
 	struct xfs_bmbt_irec	bmap;
 
-	if (!xfs_rmap_update_is_needed(mp))
+	if (!xfs_rmap_update_is_needed(mp, XFS_DATA_FORK))
 		return 0;
 
 	bmap.br_startblock = XFS_AGB_TO_FSB(mp, agno, bno);

fs/xfs/libxfs/xfs_rmap.h

@@ -206,4 +206,11 @@ int xfs_rmap_finish_one(struct xfs_trans *tp, enum xfs_rmap_intent_type type,
 		xfs_fsblock_t startblock, xfs_filblks_t blockcount,
 		xfs_exntst_t state, struct xfs_btree_cur **pcur);
 
+int xfs_rmap_find_left_neighbor(struct xfs_btree_cur *cur, xfs_agblock_t bno,
+		uint64_t owner, uint64_t offset, unsigned int flags,
+		struct xfs_rmap_irec *irec, int	*stat);
+int xfs_rmap_lookup_le_range(struct xfs_btree_cur *cur, xfs_agblock_t bno,
+		uint64_t owner, uint64_t offset, unsigned int flags,
+		struct xfs_rmap_irec *irec, int	*stat);
+
 #endif	/* __XFS_RMAP_H__ */

fs/xfs/libxfs/xfs_rmap_btree.c

@@ -35,6 +35,7 @@
 #include "xfs_cksum.h"
 #include "xfs_error.h"
 #include "xfs_extent_busy.h"
+#include "xfs_ag_resv.h"
 
 /*
  * Reverse map btree.
@@ -512,6 +513,83 @@ void
 xfs_rmapbt_compute_maxlevels(
 	struct xfs_mount		*mp)
 {
-	mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(mp,
-			mp->m_rmap_mnr, mp->m_sb.sb_agblocks);
+	/*
+	 * On a non-reflink filesystem, the maximum number of rmap
+	 * records is the number of blocks in the AG, hence the max
+	 * rmapbt height is log_$maxrecs($agblocks).  However, with
+	 * reflink each AG block can have up to 2^32 (per the refcount
+	 * record format) owners, which means that theoretically we
+	 * could face up to 2^64 rmap records.
+	 *
+	 * That effectively means that the max rmapbt height must be
+	 * XFS_BTREE_MAXLEVELS.  "Fortunately" we'll run out of AG
+	 * blocks to feed the rmapbt long before the rmapbt reaches
+	 * maximum height.  The reflink code uses ag_resv_critical to
+	 * disallow reflinking when less than 10% of the per-AG metadata
+	 * block reservation since the fallback is a regular file copy.
+	 */
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		mp->m_rmap_maxlevels = XFS_BTREE_MAXLEVELS;
+	else
+		mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(mp,
+				mp->m_rmap_mnr, mp->m_sb.sb_agblocks);
+}
+
+/* Calculate the refcount btree size for some records. */
+xfs_extlen_t
+xfs_rmapbt_calc_size(
+	struct xfs_mount	*mp,
+	unsigned long long	len)
+{
+	return xfs_btree_calc_size(mp, mp->m_rmap_mnr, len);
+}
+
+/*
+ * Calculate the maximum refcount btree size.
+ */
+xfs_extlen_t
+xfs_rmapbt_max_size(
+	struct xfs_mount	*mp)
+{
+	/* Bail out if we're uninitialized, which can happen in mkfs. */
+	if (mp->m_rmap_mxr[0] == 0)
+		return 0;
+
+	return xfs_rmapbt_calc_size(mp, mp->m_sb.sb_agblocks);
+}
+
+/*
+ * Figure out how many blocks to reserve and how many are used by this btree.
+ */
+int
+xfs_rmapbt_calc_reserves(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_extlen_t		*ask,
+	xfs_extlen_t		*used)
+{
+	struct xfs_buf		*agbp;
+	struct xfs_agf		*agf;
+	xfs_extlen_t		pool_len;
+	xfs_extlen_t		tree_len;
+	int			error;
+
+	if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
+		return 0;
+
+	/* Reserve 1% of the AG or enough for 1 block per record. */
+	pool_len = max(mp->m_sb.sb_agblocks / 100, xfs_rmapbt_max_size(mp));
+	*ask += pool_len;
+
+	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
+	if (error)
+		return error;
+
+	agf = XFS_BUF_TO_AGF(agbp);
+	tree_len = be32_to_cpu(agf->agf_rmap_blocks);
+	xfs_buf_relse(agbp);
+
+	*used += tree_len;
+
+	return error;
 }

fs/xfs/libxfs/xfs_rmap_btree.h

@@ -58,4 +58,11 @@ struct xfs_btree_cur *xfs_rmapbt_init_cursor(struct xfs_mount *mp,
 int xfs_rmapbt_maxrecs(struct xfs_mount *mp, int blocklen, int leaf);
 extern void xfs_rmapbt_compute_maxlevels(struct xfs_mount *mp);
 
+extern xfs_extlen_t xfs_rmapbt_calc_size(struct xfs_mount *mp,
+		unsigned long long len);
+extern xfs_extlen_t xfs_rmapbt_max_size(struct xfs_mount *mp);
+
+extern int xfs_rmapbt_calc_reserves(struct xfs_mount *mp,
+		xfs_agnumber_t agno, xfs_extlen_t *ask, xfs_extlen_t *used);
+
 #endif	/* __XFS_RMAP_BTREE_H__ */

fs/xfs/libxfs/xfs_sb.c

@@ -38,6 +38,8 @@
 #include "xfs_ialloc_btree.h"
 #include "xfs_log.h"
 #include "xfs_rmap_btree.h"
+#include "xfs_bmap.h"
+#include "xfs_refcount_btree.h"
 
 /*
  * Physical superblock buffer manipulations. Shared with libxfs in userspace.
@@ -737,6 +739,13 @@ xfs_sb_mount_common(
 	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
 	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
 
+	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize,
+			true);
+	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize,
+			false);
+	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
+	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
+
 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 					sbp->sb_inopblock);

fs/xfs/libxfs/xfs_shared.h

@@ -39,6 +39,7 @@ extern const struct xfs_buf_ops xfs_agf_buf_ops;
 extern const struct xfs_buf_ops xfs_agfl_buf_ops;
 extern const struct xfs_buf_ops xfs_allocbt_buf_ops;
 extern const struct xfs_buf_ops xfs_rmapbt_buf_ops;
+extern const struct xfs_buf_ops xfs_refcountbt_buf_ops;
 extern const struct xfs_buf_ops xfs_attr3_leaf_buf_ops;
 extern const struct xfs_buf_ops xfs_attr3_rmt_buf_ops;
 extern const struct xfs_buf_ops xfs_bmbt_buf_ops;
@@ -122,6 +123,7 @@ int	xfs_log_calc_minimum_size(struct xfs_mount *);
 #define	XFS_INO_REF		2
 #define	XFS_ATTR_BTREE_REF	1
 #define	XFS_DQUOT_REF		1
+#define	XFS_REFC_BTREE_REF	1
 
 /*
  * Flags for xfs_trans_ichgtime().

fs/xfs/libxfs/xfs_trans_resv.c

@@ -67,13 +67,14 @@ xfs_calc_buf_res(
  * Per-extent log reservation for the btree changes involved in freeing or
  * allocating an extent.  In classic XFS there were two trees that will be
  * modified (bnobt + cntbt).  With rmap enabled, there are three trees
- * (rmapbt).  The number of blocks reserved is based on the formula:
+ * (rmapbt).  With reflink, there are four trees (refcountbt).  The number of
+ * blocks reserved is based on the formula:
  *
  * num trees * ((2 blocks/level * max depth) - 1)
  *
  * Keep in mind that max depth is calculated separately for each type of tree.
  */
-static uint
+uint
 xfs_allocfree_log_count(
 	struct xfs_mount *mp,
 	uint		num_ops)
@@ -83,6 +84,8 @@ xfs_allocfree_log_count(
 	blocks = num_ops * 2 * (2 * mp->m_ag_maxlevels - 1);
 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
 		blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1);
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		blocks += num_ops * (2 * mp->m_refc_maxlevels - 1);
 
 	return blocks;
 }
@@ -809,11 +812,18 @@ xfs_trans_resv_calc(
 	 * require a permanent reservation on space.
 	 */
 	resp->tr_write.tr_logres = xfs_calc_write_reservation(mp);
-	resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
+	else
+		resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
 	resp->tr_write.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
 
 	resp->tr_itruncate.tr_logres = xfs_calc_itruncate_reservation(mp);
-	resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		resp->tr_itruncate.tr_logcount =
+				XFS_ITRUNCATE_LOG_COUNT_REFLINK;
+	else
+		resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
 	resp->tr_itruncate.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
 
 	resp->tr_rename.tr_logres = xfs_calc_rename_reservation(mp);
@@ -870,7 +880,10 @@ xfs_trans_resv_calc(
 	resp->tr_growrtalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
 
 	resp->tr_qm_dqalloc.tr_logres = xfs_calc_qm_dqalloc_reservation(mp);
-	resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
+	else
+		resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
 	resp->tr_qm_dqalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
 
 	/*

fs/xfs/libxfs/xfs_trans_resv.h

@@ -87,6 +87,7 @@ struct xfs_trans_resv {
 #define	XFS_DEFAULT_LOG_COUNT		1
 #define	XFS_DEFAULT_PERM_LOG_COUNT	2
 #define	XFS_ITRUNCATE_LOG_COUNT		2
+#define	XFS_ITRUNCATE_LOG_COUNT_REFLINK	8
 #define XFS_INACTIVE_LOG_COUNT		2
 #define	XFS_CREATE_LOG_COUNT		2
 #define	XFS_CREATE_TMPFILE_LOG_COUNT	2
@@ -96,11 +97,13 @@ struct xfs_trans_resv {
 #define	XFS_LINK_LOG_COUNT		2
 #define	XFS_RENAME_LOG_COUNT		2
 #define	XFS_WRITE_LOG_COUNT		2
+#define	XFS_WRITE_LOG_COUNT_REFLINK	8
 #define	XFS_ADDAFORK_LOG_COUNT		2
 #define	XFS_ATTRINVAL_LOG_COUNT		1
 #define	XFS_ATTRSET_LOG_COUNT		3
 #define	XFS_ATTRRM_LOG_COUNT		3
 
 void xfs_trans_resv_calc(struct xfs_mount *mp, struct xfs_trans_resv *resp);
+uint xfs_allocfree_log_count(struct xfs_mount *mp, uint num_ops);
 
 #endif	/* __XFS_TRANS_RESV_H__ */

fs/xfs/libxfs/xfs_trans_space.h

@@ -21,6 +21,8 @@
 /*
  * Components of space reservations.
  */
+#define XFS_MAX_CONTIG_RMAPS_PER_BLOCK(mp)    \
+		(((mp)->m_rmap_mxr[0]) - ((mp)->m_rmap_mnr[0]))
 #define XFS_MAX_CONTIG_EXTENTS_PER_BLOCK(mp)    \
 		(((mp)->m_alloc_mxr[0]) - ((mp)->m_alloc_mnr[0]))
 #define	XFS_EXTENTADD_SPACE_RES(mp,w)	(XFS_BM_MAXLEVELS(mp,w) - 1)
@@ -28,6 +30,13 @@
 	(((b + XFS_MAX_CONTIG_EXTENTS_PER_BLOCK(mp) - 1) / \
 	  XFS_MAX_CONTIG_EXTENTS_PER_BLOCK(mp)) * \
 	  XFS_EXTENTADD_SPACE_RES(mp,w))
+#define XFS_SWAP_RMAP_SPACE_RES(mp,b,w)\
+	(((b + XFS_MAX_CONTIG_EXTENTS_PER_BLOCK(mp) - 1) / \
+	  XFS_MAX_CONTIG_EXTENTS_PER_BLOCK(mp)) * \
+	  XFS_EXTENTADD_SPACE_RES(mp,w) + \
+	 ((b + XFS_MAX_CONTIG_RMAPS_PER_BLOCK(mp) - 1) / \
+	  XFS_MAX_CONTIG_RMAPS_PER_BLOCK(mp)) * \
+	  (mp)->m_rmap_maxlevels)
 #define	XFS_DAENTER_1B(mp,w)	\
 	((w) == XFS_DATA_FORK ? (mp)->m_dir_geo->fsbcount : 1)
 #define	XFS_DAENTER_DBS(mp,w)	\

fs/xfs/libxfs/xfs_types.h

@@ -90,6 +90,7 @@ typedef __int64_t	xfs_sfiloff_t;	/* signed block number in a file */
  */
 #define	XFS_DATA_FORK	0
 #define	XFS_ATTR_FORK	1
+#define	XFS_COW_FORK	2
 
 /*
  * Min numbers of data/attr fork btree root pointers.
@@ -109,7 +110,7 @@ typedef enum {
 
 typedef enum {
 	XFS_BTNUM_BNOi, XFS_BTNUM_CNTi, XFS_BTNUM_RMAPi, XFS_BTNUM_BMAPi,
-	XFS_BTNUM_INOi, XFS_BTNUM_FINOi, XFS_BTNUM_MAX
+	XFS_BTNUM_INOi, XFS_BTNUM_FINOi, XFS_BTNUM_REFCi, XFS_BTNUM_MAX
 } xfs_btnum_t;
 
 struct xfs_name {

fs/xfs/xfs_aops.c

@@ -31,6 +31,7 @@
 #include "xfs_bmap.h"
 #include "xfs_bmap_util.h"
 #include "xfs_bmap_btree.h"
+#include "xfs_reflink.h"
 #include <linux/gfp.h>
 #include <linux/mpage.h>
 #include <linux/pagevec.h>
@@ -39,6 +40,7 @@
 /* flags for direct write completions */
 #define XFS_DIO_FLAG_UNWRITTEN	(1 << 0)
 #define XFS_DIO_FLAG_APPEND	(1 << 1)
+#define XFS_DIO_FLAG_COW	(1 << 2)
 
 /*
  * structure owned by writepages passed to individual writepage calls
@@ -286,6 +288,25 @@ xfs_end_io(
 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 		error = -EIO;
 
+	/*
+	 * For a CoW extent, we need to move the mapping from the CoW fork
+	 * to the data fork.  If instead an error happened, just dump the
+	 * new blocks.
+	 */
+	if (ioend->io_type == XFS_IO_COW) {
+		if (error)
+			goto done;
+		if (ioend->io_bio->bi_error) {
+			error = xfs_reflink_cancel_cow_range(ip,
+					ioend->io_offset, ioend->io_size);
+			goto done;
+		}
+		error = xfs_reflink_end_cow(ip, ioend->io_offset,
+				ioend->io_size);
+		if (error)
+			goto done;
+	}
+
 	/*
 	 * For unwritten extents we need to issue transactions to convert a
 	 * range to normal written extens after the data I/O has finished.
@@ -301,7 +322,8 @@ xfs_end_io(
 	} else if (ioend->io_append_trans) {
 		error = xfs_setfilesize_ioend(ioend, error);
 	} else {
-		ASSERT(!xfs_ioend_is_append(ioend));
+		ASSERT(!xfs_ioend_is_append(ioend) ||
+		       ioend->io_type == XFS_IO_COW);
 	}
 
 done:
@@ -315,7 +337,7 @@ xfs_end_bio(
 	struct xfs_ioend	*ioend = bio->bi_private;
 	struct xfs_mount	*mp = XFS_I(ioend->io_inode)->i_mount;
 
-	if (ioend->io_type == XFS_IO_UNWRITTEN)
+	if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
 		queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
 	else if (ioend->io_append_trans)
 		queue_work(mp->m_data_workqueue, &ioend->io_work);
@@ -341,6 +363,7 @@ xfs_map_blocks(
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return -EIO;
 
+	ASSERT(type != XFS_IO_COW);
 	if (type == XFS_IO_UNWRITTEN)
 		bmapi_flags |= XFS_BMAPI_IGSTATE;
 
@@ -355,6 +378,13 @@ xfs_map_blocks(
 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
 				imap, &nimaps, bmapi_flags);
+	/*
+	 * Truncate an overwrite extent if there's a pending CoW
+	 * reservation before the end of this extent.  This forces us
+	 * to come back to writepage to take care of the CoW.
+	 */
+	if (nimaps && type == XFS_IO_OVERWRITE)
+		xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 	if (error)
@@ -362,7 +392,8 @@ xfs_map_blocks(
 
 	if (type == XFS_IO_DELALLOC &&
 	    (!nimaps || isnullstartblock(imap->br_startblock))) {
-		error = xfs_iomap_write_allocate(ip, offset, imap);
+		error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
+				imap);
 		if (!error)
 			trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
 		return error;
@@ -737,6 +768,56 @@ out_invalidate:
 	return;
 }
 
+static int
+xfs_map_cow(
+	struct xfs_writepage_ctx *wpc,
+	struct inode		*inode,
+	loff_t			offset,
+	unsigned int		*new_type)
+{
+	struct xfs_inode	*ip = XFS_I(inode);
+	struct xfs_bmbt_irec	imap;
+	bool			is_cow = false, need_alloc = false;
+	int			error;
+
+	/*
+	 * If we already have a valid COW mapping keep using it.
+	 */
+	if (wpc->io_type == XFS_IO_COW) {
+		wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
+		if (wpc->imap_valid) {
+			*new_type = XFS_IO_COW;
+			return 0;
+		}
+	}
+
+	/*
+	 * Else we need to check if there is a COW mapping at this offset.
+	 */
+	xfs_ilock(ip, XFS_ILOCK_SHARED);
+	is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, &need_alloc);
+	xfs_iunlock(ip, XFS_ILOCK_SHARED);
+
+	if (!is_cow)
+		return 0;
+
+	/*
+	 * And if the COW mapping has a delayed extent here we need to
+	 * allocate real space for it now.
+	 */
+	if (need_alloc) {
+		error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
+				&imap);
+		if (error)
+			return error;
+	}
+
+	wpc->io_type = *new_type = XFS_IO_COW;
+	wpc->imap_valid = true;
+	wpc->imap = imap;
+	return 0;
+}
+
 /*
  * We implement an immediate ioend submission policy here to avoid needing to
  * chain multiple ioends and hence nest mempool allocations which can violate
@@ -769,6 +850,7 @@ xfs_writepage_map(
 	int			error = 0;
 	int			count = 0;
 	int			uptodate = 1;
+	unsigned int		new_type;
 
 	bh = head = page_buffers(page);
 	offset = page_offset(page);
@@ -789,22 +871,13 @@ xfs_writepage_map(
 			continue;
 		}
 
-		if (buffer_unwritten(bh)) {
-			if (wpc->io_type != XFS_IO_UNWRITTEN) {
-				wpc->io_type = XFS_IO_UNWRITTEN;
-				wpc->imap_valid = false;
-			}
-		} else if (buffer_delay(bh)) {
-			if (wpc->io_type != XFS_IO_DELALLOC) {
-				wpc->io_type = XFS_IO_DELALLOC;
-				wpc->imap_valid = false;
-			}
-		} else if (buffer_uptodate(bh)) {
-			if (wpc->io_type != XFS_IO_OVERWRITE) {
-				wpc->io_type = XFS_IO_OVERWRITE;
-				wpc->imap_valid = false;
-			}
-		} else {
+		if (buffer_unwritten(bh))
+			new_type = XFS_IO_UNWRITTEN;
+		else if (buffer_delay(bh))
+			new_type = XFS_IO_DELALLOC;
+		else if (buffer_uptodate(bh))
+			new_type = XFS_IO_OVERWRITE;
+		else {
 			if (PageUptodate(page))
 				ASSERT(buffer_mapped(bh));
 			/*
@@ -817,6 +890,17 @@ xfs_writepage_map(
 			continue;
 		}
 
+		if (xfs_is_reflink_inode(XFS_I(inode))) {
+			error = xfs_map_cow(wpc, inode, offset, &new_type);
+			if (error)
+				goto out;
+		}
+
+		if (wpc->io_type != new_type) {
+			wpc->io_type = new_type;
+			wpc->imap_valid = false;
+		}
+
 		if (wpc->imap_valid)
 			wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
 							 offset);
@@ -1107,18 +1191,24 @@ xfs_map_direct(
 	struct inode		*inode,
 	struct buffer_head	*bh_result,
 	struct xfs_bmbt_irec	*imap,
-	xfs_off_t		offset)
+	xfs_off_t		offset,
+	bool			is_cow)
 {
 	uintptr_t		*flags = (uintptr_t *)&bh_result->b_private;
 	xfs_off_t		size = bh_result->b_size;
 
 	trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
-		ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap);
+		ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : is_cow ? XFS_IO_COW :
+		XFS_IO_OVERWRITE, imap);
 
 	if (ISUNWRITTEN(imap)) {
 		*flags |= XFS_DIO_FLAG_UNWRITTEN;
 		set_buffer_defer_completion(bh_result);
-	} else if (offset + size > i_size_read(inode) || offset + size < 0) {
+	} else if (is_cow) {
+		*flags |= XFS_DIO_FLAG_COW;
+		set_buffer_defer_completion(bh_result);
+	}
+	if (offset + size > i_size_read(inode) || offset + size < 0) {
 		*flags |= XFS_DIO_FLAG_APPEND;
 		set_buffer_defer_completion(bh_result);
 	}
@@ -1164,6 +1254,44 @@ xfs_map_trim_size(
 	bh_result->b_size = mapping_size;
 }
 
+/* Bounce unaligned directio writes to the page cache. */
+static int
+xfs_bounce_unaligned_dio_write(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		offset_fsb,
+	struct xfs_bmbt_irec	*imap)
+{
+	struct xfs_bmbt_irec	irec;
+	xfs_fileoff_t		delta;
+	bool			shared;
+	bool			x;
+	int			error;
+
+	irec = *imap;
+	if (offset_fsb > irec.br_startoff) {
+		delta = offset_fsb - irec.br_startoff;
+		irec.br_blockcount -= delta;
+		irec.br_startblock += delta;
+		irec.br_startoff = offset_fsb;
+	}
+	error = xfs_reflink_trim_around_shared(ip, &irec, &shared, &x);
+	if (error)
+		return error;
+
+	/*
+	 * We're here because we're trying to do a directio write to a
+	 * region that isn't aligned to a filesystem block.  If any part
+	 * of the extent is shared, fall back to buffered mode to handle
+	 * the RMW.  This is done by returning -EREMCHG ("remote addr
+	 * changed"), which is caught further up the call stack.
+	 */
+	if (shared) {
+		trace_xfs_reflink_bounce_dio_write(ip, imap);
+		return -EREMCHG;
+	}
+	return 0;
+}
+
 STATIC int
 __xfs_get_blocks(
 	struct inode		*inode,
@@ -1183,6 +1311,8 @@ __xfs_get_blocks(
 	xfs_off_t		offset;
 	ssize_t			size;
 	int			new = 0;
+	bool			is_cow = false;
+	bool			need_alloc = false;
 
 	BUG_ON(create && !direct);
 
@@ -1208,8 +1338,26 @@ __xfs_get_blocks(
 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 
-	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
-				&imap, &nimaps, XFS_BMAPI_ENTIRE);
+	if (create && direct && xfs_is_reflink_inode(ip))
+		is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap,
+					&need_alloc);
+	if (!is_cow) {
+		error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
+					&imap, &nimaps, XFS_BMAPI_ENTIRE);
+		/*
+		 * Truncate an overwrite extent if there's a pending CoW
+		 * reservation before the end of this extent.  This
+		 * forces us to come back to get_blocks to take care of
+		 * the CoW.
+		 */
+		if (create && direct && nimaps &&
+		    imap.br_startblock != HOLESTARTBLOCK &&
+		    imap.br_startblock != DELAYSTARTBLOCK &&
+		    !ISUNWRITTEN(&imap))
+			xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb,
+					&imap);
+	}
+	ASSERT(!need_alloc);
 	if (error)
 		goto out_unlock;
 
@@ -1261,6 +1409,13 @@ __xfs_get_blocks(
 	if (imap.br_startblock != HOLESTARTBLOCK &&
 	    imap.br_startblock != DELAYSTARTBLOCK &&
 	    (create || !ISUNWRITTEN(&imap))) {
+		if (create && direct && !is_cow) {
+			error = xfs_bounce_unaligned_dio_write(ip, offset_fsb,
+					&imap);
+			if (error)
+				return error;
+		}
+
 		xfs_map_buffer(inode, bh_result, &imap, offset);
 		if (ISUNWRITTEN(&imap))
 			set_buffer_unwritten(bh_result);
@@ -1269,7 +1424,8 @@ __xfs_get_blocks(
 			if (dax_fault)
 				ASSERT(!ISUNWRITTEN(&imap));
 			else
-				xfs_map_direct(inode, bh_result, &imap, offset);
+				xfs_map_direct(inode, bh_result, &imap, offset,
+						is_cow);
 		}
 	}
 
@@ -1391,11 +1547,14 @@ xfs_end_io_direct_write(
 		i_size_write(inode, offset + size);
 	spin_unlock(&ip->i_flags_lock);
 
+	if (flags & XFS_DIO_FLAG_COW)
+		error = xfs_reflink_end_cow(ip, offset, size);
 	if (flags & XFS_DIO_FLAG_UNWRITTEN) {
 		trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
 
 		error = xfs_iomap_write_unwritten(ip, offset, size);
-	} else if (flags & XFS_DIO_FLAG_APPEND) {
+	}
+	if (flags & XFS_DIO_FLAG_APPEND) {
 		trace_xfs_end_io_direct_write_append(ip, offset, size);
 
 		error = xfs_setfilesize(ip, offset, size);
@@ -1425,6 +1584,17 @@ xfs_vm_bmap(
 
 	trace_xfs_vm_bmap(XFS_I(inode));
 	xfs_ilock(ip, XFS_IOLOCK_SHARED);
+
+	/*
+	 * The swap code (ab-)uses ->bmap to get a block mapping and then
+	 * bypasseѕ the file system for actual I/O.  We really can't allow
+	 * that on reflinks inodes, so we have to skip out here.  And yes,
+	 * 0 is the magic code for a bmap error..
+	 */
+	if (xfs_is_reflink_inode(ip)) {
+		xfs_iunlock(ip, XFS_IOLOCK_SHARED);
+		return 0;
+	}
 	filemap_write_and_wait(mapping);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 	return generic_block_bmap(mapping, block, xfs_get_blocks);

fs/xfs/xfs_aops.h

@@ -28,13 +28,15 @@ enum {
 	XFS_IO_DELALLOC,	/* covers delalloc region */
 	XFS_IO_UNWRITTEN,	/* covers allocated but uninitialized data */
 	XFS_IO_OVERWRITE,	/* covers already allocated extent */
+	XFS_IO_COW,		/* covers copy-on-write extent */
 };
 
 #define XFS_IO_TYPES \
 	{ XFS_IO_INVALID,		"invalid" }, \
 	{ XFS_IO_DELALLOC,		"delalloc" }, \
 	{ XFS_IO_UNWRITTEN,		"unwritten" }, \
-	{ XFS_IO_OVERWRITE,		"overwrite" }
+	{ XFS_IO_OVERWRITE,		"overwrite" }, \
+	{ XFS_IO_COW,			"CoW" }
 
 /*
  * Structure for buffered I/O completions.

fs/xfs/xfs_bmap_item.c

@@ -0,0 +1,508 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_bit.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_inode.h"
+#include "xfs_trans.h"
+#include "xfs_trans_priv.h"
+#include "xfs_buf_item.h"
+#include "xfs_bmap_item.h"
+#include "xfs_log.h"
+#include "xfs_bmap.h"
+#include "xfs_icache.h"
+#include "xfs_trace.h"
+
+
+kmem_zone_t	*xfs_bui_zone;
+kmem_zone_t	*xfs_bud_zone;
+
+static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
+{
+	return container_of(lip, struct xfs_bui_log_item, bui_item);
+}
+
+void
+xfs_bui_item_free(
+	struct xfs_bui_log_item	*buip)
+{
+	kmem_zone_free(xfs_bui_zone, buip);
+}
+
+STATIC void
+xfs_bui_item_size(
+	struct xfs_log_item	*lip,
+	int			*nvecs,
+	int			*nbytes)
+{
+	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
+
+	*nvecs += 1;
+	*nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
+}
+
+/*
+ * This is called to fill in the vector of log iovecs for the
+ * given bui log item. We use only 1 iovec, and we point that
+ * at the bui_log_format structure embedded in the bui item.
+ * It is at this point that we assert that all of the extent
+ * slots in the bui item have been filled.
+ */
+STATIC void
+xfs_bui_item_format(
+	struct xfs_log_item	*lip,
+	struct xfs_log_vec	*lv)
+{
+	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
+
+	ASSERT(atomic_read(&buip->bui_next_extent) ==
+			buip->bui_format.bui_nextents);
+
+	buip->bui_format.bui_type = XFS_LI_BUI;
+	buip->bui_format.bui_size = 1;
+
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
+			xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
+}
+
+/*
+ * Pinning has no meaning for an bui item, so just return.
+ */
+STATIC void
+xfs_bui_item_pin(
+	struct xfs_log_item	*lip)
+{
+}
+
+/*
+ * The unpin operation is the last place an BUI is manipulated in the log. It is
+ * either inserted in the AIL or aborted in the event of a log I/O error. In
+ * either case, the BUI transaction has been successfully committed to make it
+ * this far. Therefore, we expect whoever committed the BUI to either construct
+ * and commit the BUD or drop the BUD's reference in the event of error. Simply
+ * drop the log's BUI reference now that the log is done with it.
+ */
+STATIC void
+xfs_bui_item_unpin(
+	struct xfs_log_item	*lip,
+	int			remove)
+{
+	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
+
+	xfs_bui_release(buip);
+}
+
+/*
+ * BUI items have no locking or pushing.  However, since BUIs are pulled from
+ * the AIL when their corresponding BUDs are committed to disk, their situation
+ * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
+ * will eventually flush the log.  This should help in getting the BUI out of
+ * the AIL.
+ */
+STATIC uint
+xfs_bui_item_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
+{
+	return XFS_ITEM_PINNED;
+}
+
+/*
+ * The BUI has been either committed or aborted if the transaction has been
+ * cancelled. If the transaction was cancelled, an BUD isn't going to be
+ * constructed and thus we free the BUI here directly.
+ */
+STATIC void
+xfs_bui_item_unlock(
+	struct xfs_log_item	*lip)
+{
+	if (lip->li_flags & XFS_LI_ABORTED)
+		xfs_bui_item_free(BUI_ITEM(lip));
+}
+
+/*
+ * The BUI is logged only once and cannot be moved in the log, so simply return
+ * the lsn at which it's been logged.
+ */
+STATIC xfs_lsn_t
+xfs_bui_item_committed(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+	return lsn;
+}
+
+/*
+ * The BUI dependency tracking op doesn't do squat.  It can't because
+ * it doesn't know where the free extent is coming from.  The dependency
+ * tracking has to be handled by the "enclosing" metadata object.  For
+ * example, for inodes, the inode is locked throughout the extent freeing
+ * so the dependency should be recorded there.
+ */
+STATIC void
+xfs_bui_item_committing(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+}
+
+/*
+ * This is the ops vector shared by all bui log items.
+ */
+static const struct xfs_item_ops xfs_bui_item_ops = {
+	.iop_size	= xfs_bui_item_size,
+	.iop_format	= xfs_bui_item_format,
+	.iop_pin	= xfs_bui_item_pin,
+	.iop_unpin	= xfs_bui_item_unpin,
+	.iop_unlock	= xfs_bui_item_unlock,
+	.iop_committed	= xfs_bui_item_committed,
+	.iop_push	= xfs_bui_item_push,
+	.iop_committing = xfs_bui_item_committing,
+};
+
+/*
+ * Allocate and initialize an bui item with the given number of extents.
+ */
+struct xfs_bui_log_item *
+xfs_bui_init(
+	struct xfs_mount		*mp)
+
+{
+	struct xfs_bui_log_item		*buip;
+
+	buip = kmem_zone_zalloc(xfs_bui_zone, KM_SLEEP);
+
+	xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
+	buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
+	buip->bui_format.bui_id = (uintptr_t)(void *)buip;
+	atomic_set(&buip->bui_next_extent, 0);
+	atomic_set(&buip->bui_refcount, 2);
+
+	return buip;
+}
+
+/*
+ * Freeing the BUI requires that we remove it from the AIL if it has already
+ * been placed there. However, the BUI may not yet have been placed in the AIL
+ * when called by xfs_bui_release() from BUD processing due to the ordering of
+ * committed vs unpin operations in bulk insert operations. Hence the reference
+ * count to ensure only the last caller frees the BUI.
+ */
+void
+xfs_bui_release(
+	struct xfs_bui_log_item	*buip)
+{
+	if (atomic_dec_and_test(&buip->bui_refcount)) {
+		xfs_trans_ail_remove(&buip->bui_item, SHUTDOWN_LOG_IO_ERROR);
+		xfs_bui_item_free(buip);
+	}
+}
+
+static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
+{
+	return container_of(lip, struct xfs_bud_log_item, bud_item);
+}
+
+STATIC void
+xfs_bud_item_size(
+	struct xfs_log_item	*lip,
+	int			*nvecs,
+	int			*nbytes)
+{
+	*nvecs += 1;
+	*nbytes += sizeof(struct xfs_bud_log_format);
+}
+
+/*
+ * This is called to fill in the vector of log iovecs for the
+ * given bud log item. We use only 1 iovec, and we point that
+ * at the bud_log_format structure embedded in the bud item.
+ * It is at this point that we assert that all of the extent
+ * slots in the bud item have been filled.
+ */
+STATIC void
+xfs_bud_item_format(
+	struct xfs_log_item	*lip,
+	struct xfs_log_vec	*lv)
+{
+	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
+
+	budp->bud_format.bud_type = XFS_LI_BUD;
+	budp->bud_format.bud_size = 1;
+
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
+			sizeof(struct xfs_bud_log_format));
+}
+
+/*
+ * Pinning has no meaning for an bud item, so just return.
+ */
+STATIC void
+xfs_bud_item_pin(
+	struct xfs_log_item	*lip)
+{
+}
+
+/*
+ * Since pinning has no meaning for an bud item, unpinning does
+ * not either.
+ */
+STATIC void
+xfs_bud_item_unpin(
+	struct xfs_log_item	*lip,
+	int			remove)
+{
+}
+
+/*
+ * There isn't much you can do to push on an bud item.  It is simply stuck
+ * waiting for the log to be flushed to disk.
+ */
+STATIC uint
+xfs_bud_item_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
+{
+	return XFS_ITEM_PINNED;
+}
+
+/*
+ * The BUD is either committed or aborted if the transaction is cancelled. If
+ * the transaction is cancelled, drop our reference to the BUI and free the
+ * BUD.
+ */
+STATIC void
+xfs_bud_item_unlock(
+	struct xfs_log_item	*lip)
+{
+	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
+
+	if (lip->li_flags & XFS_LI_ABORTED) {
+		xfs_bui_release(budp->bud_buip);
+		kmem_zone_free(xfs_bud_zone, budp);
+	}
+}
+
+/*
+ * When the bud item is committed to disk, all we need to do is delete our
+ * reference to our partner bui item and then free ourselves. Since we're
+ * freeing ourselves we must return -1 to keep the transaction code from
+ * further referencing this item.
+ */
+STATIC xfs_lsn_t
+xfs_bud_item_committed(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
+
+	/*
+	 * Drop the BUI reference regardless of whether the BUD has been
+	 * aborted. Once the BUD transaction is constructed, it is the sole
+	 * responsibility of the BUD to release the BUI (even if the BUI is
+	 * aborted due to log I/O error).
+	 */
+	xfs_bui_release(budp->bud_buip);
+	kmem_zone_free(xfs_bud_zone, budp);
+
+	return (xfs_lsn_t)-1;
+}
+
+/*
+ * The BUD dependency tracking op doesn't do squat.  It can't because
+ * it doesn't know where the free extent is coming from.  The dependency
+ * tracking has to be handled by the "enclosing" metadata object.  For
+ * example, for inodes, the inode is locked throughout the extent freeing
+ * so the dependency should be recorded there.
+ */
+STATIC void
+xfs_bud_item_committing(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+}
+
+/*
+ * This is the ops vector shared by all bud log items.
+ */
+static const struct xfs_item_ops xfs_bud_item_ops = {
+	.iop_size	= xfs_bud_item_size,
+	.iop_format	= xfs_bud_item_format,
+	.iop_pin	= xfs_bud_item_pin,
+	.iop_unpin	= xfs_bud_item_unpin,
+	.iop_unlock	= xfs_bud_item_unlock,
+	.iop_committed	= xfs_bud_item_committed,
+	.iop_push	= xfs_bud_item_push,
+	.iop_committing = xfs_bud_item_committing,
+};
+
+/*
+ * Allocate and initialize an bud item with the given number of extents.
+ */
+struct xfs_bud_log_item *
+xfs_bud_init(
+	struct xfs_mount		*mp,
+	struct xfs_bui_log_item		*buip)
+
+{
+	struct xfs_bud_log_item	*budp;
+
+	budp = kmem_zone_zalloc(xfs_bud_zone, KM_SLEEP);
+	xfs_log_item_init(mp, &budp->bud_item, XFS_LI_BUD, &xfs_bud_item_ops);
+	budp->bud_buip = buip;
+	budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
+
+	return budp;
+}
+
+/*
+ * Process a bmap update intent item that was recovered from the log.
+ * We need to update some inode's bmbt.
+ */
+int
+xfs_bui_recover(
+	struct xfs_mount		*mp,
+	struct xfs_bui_log_item		*buip)
+{
+	int				error = 0;
+	unsigned int			bui_type;
+	struct xfs_map_extent		*bmap;
+	xfs_fsblock_t			startblock_fsb;
+	xfs_fsblock_t			inode_fsb;
+	bool				op_ok;
+	struct xfs_bud_log_item		*budp;
+	enum xfs_bmap_intent_type	type;
+	int				whichfork;
+	xfs_exntst_t			state;
+	struct xfs_trans		*tp;
+	struct xfs_inode		*ip = NULL;
+	struct xfs_defer_ops		dfops;
+	xfs_fsblock_t			firstfsb;
+
+	ASSERT(!test_bit(XFS_BUI_RECOVERED, &buip->bui_flags));
+
+	/* Only one mapping operation per BUI... */
+	if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
+		set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
+		xfs_bui_release(buip);
+		return -EIO;
+	}
+
+	/*
+	 * First check the validity of the extent described by the
+	 * BUI.  If anything is bad, then toss the BUI.
+	 */
+	bmap = &buip->bui_format.bui_extents[0];
+	startblock_fsb = XFS_BB_TO_FSB(mp,
+			   XFS_FSB_TO_DADDR(mp, bmap->me_startblock));
+	inode_fsb = XFS_BB_TO_FSB(mp, XFS_FSB_TO_DADDR(mp,
+			XFS_INO_TO_FSB(mp, bmap->me_owner)));
+	switch (bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
+	case XFS_BMAP_MAP:
+	case XFS_BMAP_UNMAP:
+		op_ok = true;
+		break;
+	default:
+		op_ok = false;
+		break;
+	}
+	if (!op_ok || startblock_fsb == 0 ||
+	    bmap->me_len == 0 ||
+	    inode_fsb == 0 ||
+	    startblock_fsb >= mp->m_sb.sb_dblocks ||
+	    bmap->me_len >= mp->m_sb.sb_agblocks ||
+	    inode_fsb >= mp->m_sb.sb_dblocks ||
+	    (bmap->me_flags & ~XFS_BMAP_EXTENT_FLAGS)) {
+		/*
+		 * This will pull the BUI from the AIL and
+		 * free the memory associated with it.
+		 */
+		set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
+		xfs_bui_release(buip);
+		return -EIO;
+	}
+
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
+	if (error)
+		return error;
+	budp = xfs_trans_get_bud(tp, buip);
+
+	/* Grab the inode. */
+	error = xfs_iget(mp, tp, bmap->me_owner, 0, XFS_ILOCK_EXCL, &ip);
+	if (error)
+		goto err_inode;
+
+	if (VFS_I(ip)->i_nlink == 0)
+		xfs_iflags_set(ip, XFS_IRECOVERY);
+	xfs_defer_init(&dfops, &firstfsb);
+
+	/* Process deferred bmap item. */
+	state = (bmap->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
+			XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
+	whichfork = (bmap->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
+			XFS_ATTR_FORK : XFS_DATA_FORK;
+	bui_type = bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
+	switch (bui_type) {
+	case XFS_BMAP_MAP:
+	case XFS_BMAP_UNMAP:
+		type = bui_type;
+		break;
+	default:
+		error = -EFSCORRUPTED;
+		goto err_dfops;
+	}
+	xfs_trans_ijoin(tp, ip, 0);
+
+	error = xfs_trans_log_finish_bmap_update(tp, budp, &dfops, type,
+			ip, whichfork, bmap->me_startoff,
+			bmap->me_startblock, bmap->me_len,
+			state);
+	if (error)
+		goto err_dfops;
+
+	/* Finish transaction, free inodes. */
+	error = xfs_defer_finish(&tp, &dfops, NULL);
+	if (error)
+		goto err_dfops;
+
+	set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
+	error = xfs_trans_commit(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	IRELE(ip);
+
+	return error;
+
+err_dfops:
+	xfs_defer_cancel(&dfops);
+err_inode:
+	xfs_trans_cancel(tp);
+	if (ip) {
+		xfs_iunlock(ip, XFS_ILOCK_EXCL);
+		IRELE(ip);
+	}
+	return error;
+}

fs/xfs/xfs_bmap_item.h

@@ -0,0 +1,98 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef	__XFS_BMAP_ITEM_H__
+#define	__XFS_BMAP_ITEM_H__
+
+/*
+ * There are (currently) two pairs of bmap btree redo item types: map & unmap.
+ * The common abbreviations for these are BUI (bmap update intent) and BUD
+ * (bmap update done).  The redo item type is encoded in the flags field of
+ * each xfs_map_extent.
+ *
+ * *I items should be recorded in the *first* of a series of rolled
+ * transactions, and the *D items should be recorded in the same transaction
+ * that records the associated bmbt updates.
+ *
+ * Should the system crash after the commit of the first transaction but
+ * before the commit of the final transaction in a series, log recovery will
+ * use the redo information recorded by the intent items to replay the
+ * bmbt metadata updates in the non-first transaction.
+ */
+
+/* kernel only BUI/BUD definitions */
+
+struct xfs_mount;
+struct kmem_zone;
+
+/*
+ * Max number of extents in fast allocation path.
+ */
+#define	XFS_BUI_MAX_FAST_EXTENTS	1
+
+/*
+ * Define BUI flag bits. Manipulated by set/clear/test_bit operators.
+ */
+#define	XFS_BUI_RECOVERED		1
+
+/*
+ * This is the "bmap update intent" log item.  It is used to log the fact that
+ * some reverse mappings need to change.  It is used in conjunction with the
+ * "bmap update done" log item described below.
+ *
+ * These log items follow the same rules as struct xfs_efi_log_item; see the
+ * comments about that structure (in xfs_extfree_item.h) for more details.
+ */
+struct xfs_bui_log_item {
+	struct xfs_log_item		bui_item;
+	atomic_t			bui_refcount;
+	atomic_t			bui_next_extent;
+	unsigned long			bui_flags;	/* misc flags */
+	struct xfs_bui_log_format	bui_format;
+};
+
+static inline size_t
+xfs_bui_log_item_sizeof(
+	unsigned int		nr)
+{
+	return offsetof(struct xfs_bui_log_item, bui_format) +
+			xfs_bui_log_format_sizeof(nr);
+}
+
+/*
+ * This is the "bmap update done" log item.  It is used to log the fact that
+ * some bmbt updates mentioned in an earlier bui item have been performed.
+ */
+struct xfs_bud_log_item {
+	struct xfs_log_item		bud_item;
+	struct xfs_bui_log_item		*bud_buip;
+	struct xfs_bud_log_format	bud_format;
+};
+
+extern struct kmem_zone	*xfs_bui_zone;
+extern struct kmem_zone	*xfs_bud_zone;
+
+struct xfs_bui_log_item *xfs_bui_init(struct xfs_mount *);
+struct xfs_bud_log_item *xfs_bud_init(struct xfs_mount *,
+		struct xfs_bui_log_item *);
+void xfs_bui_item_free(struct xfs_bui_log_item *);
+void xfs_bui_release(struct xfs_bui_log_item *);
+int xfs_bui_recover(struct xfs_mount *mp, struct xfs_bui_log_item *buip);
+
+#endif	/* __XFS_BMAP_ITEM_H__ */

fs/xfs/xfs_bmap_util.c

@@ -42,6 +42,9 @@
 #include "xfs_icache.h"
 #include "xfs_log.h"
 #include "xfs_rmap_btree.h"
+#include "xfs_iomap.h"
+#include "xfs_reflink.h"
+#include "xfs_refcount.h"
 
 /* Kernel only BMAP related definitions and functions */
 
@@ -389,11 +392,13 @@ xfs_bmap_count_blocks(
 STATIC int
 xfs_getbmapx_fix_eof_hole(
 	xfs_inode_t		*ip,		/* xfs incore inode pointer */
+	int			whichfork,
 	struct getbmapx		*out,		/* output structure */
 	int			prealloced,	/* this is a file with
 						 * preallocated data space */
 	__int64_t		end,		/* last block requested */
-	xfs_fsblock_t		startblock)
+	xfs_fsblock_t		startblock,
+	bool			moretocome)
 {
 	__int64_t		fixlen;
 	xfs_mount_t		*mp;		/* file system mount point */
@@ -418,8 +423,9 @@ xfs_getbmapx_fix_eof_hole(
 		else
 			out->bmv_block = xfs_fsb_to_db(ip, startblock);
 		fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
-		ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
-		if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
+		ifp = XFS_IFORK_PTR(ip, whichfork);
+		if (!moretocome &&
+		    xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
 		   (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
 			out->bmv_oflags |= BMV_OF_LAST;
 	}
@@ -427,6 +433,81 @@ xfs_getbmapx_fix_eof_hole(
 	return 1;
 }
 
+/* Adjust the reported bmap around shared/unshared extent transitions. */
+STATIC int
+xfs_getbmap_adjust_shared(
+	struct xfs_inode		*ip,
+	int				whichfork,
+	struct xfs_bmbt_irec		*map,
+	struct getbmapx			*out,
+	struct xfs_bmbt_irec		*next_map)
+{
+	struct xfs_mount		*mp = ip->i_mount;
+	xfs_agnumber_t			agno;
+	xfs_agblock_t			agbno;
+	xfs_agblock_t			ebno;
+	xfs_extlen_t			elen;
+	xfs_extlen_t			nlen;
+	int				error;
+
+	next_map->br_startblock = NULLFSBLOCK;
+	next_map->br_startoff = NULLFILEOFF;
+	next_map->br_blockcount = 0;
+
+	/* Only written data blocks can be shared. */
+	if (!xfs_is_reflink_inode(ip) || whichfork != XFS_DATA_FORK ||
+	    map->br_startblock == DELAYSTARTBLOCK ||
+	    map->br_startblock == HOLESTARTBLOCK ||
+	    ISUNWRITTEN(map))
+		return 0;
+
+	agno = XFS_FSB_TO_AGNO(mp, map->br_startblock);
+	agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock);
+	error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount,
+			&ebno, &elen, true);
+	if (error)
+		return error;
+
+	if (ebno == NULLAGBLOCK) {
+		/* No shared blocks at all. */
+		return 0;
+	} else if (agbno == ebno) {
+		/*
+		 * Shared extent at (agbno, elen).  Shrink the reported
+		 * extent length and prepare to move the start of map[i]
+		 * to agbno+elen, with the aim of (re)formatting the new
+		 * map[i] the next time through the inner loop.
+		 */
+		out->bmv_length = XFS_FSB_TO_BB(mp, elen);
+		out->bmv_oflags |= BMV_OF_SHARED;
+		if (elen != map->br_blockcount) {
+			*next_map = *map;
+			next_map->br_startblock += elen;
+			next_map->br_startoff += elen;
+			next_map->br_blockcount -= elen;
+		}
+		map->br_blockcount -= elen;
+	} else {
+		/*
+		 * There's an unshared extent (agbno, ebno - agbno)
+		 * followed by shared extent at (ebno, elen).  Shrink
+		 * the reported extent length to cover only the unshared
+		 * extent and prepare to move up the start of map[i] to
+		 * ebno, with the aim of (re)formatting the new map[i]
+		 * the next time through the inner loop.
+		 */
+		*next_map = *map;
+		nlen = ebno - agbno;
+		out->bmv_length = XFS_FSB_TO_BB(mp, nlen);
+		next_map->br_startblock += nlen;
+		next_map->br_startoff += nlen;
+		next_map->br_blockcount -= nlen;
+		map->br_blockcount -= nlen;
+	}
+
+	return 0;
+}
+
 /*
  * Get inode's extents as described in bmv, and format for output.
  * Calls formatter to fill the user's buffer until all extents
@@ -459,12 +540,28 @@ xfs_getbmap(
 	int			iflags;		/* interface flags */
 	int			bmapi_flags;	/* flags for xfs_bmapi */
 	int			cur_ext = 0;
+	struct xfs_bmbt_irec	inject_map;
 
 	mp = ip->i_mount;
 	iflags = bmv->bmv_iflags;
-	whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
 
-	if (whichfork == XFS_ATTR_FORK) {
+#ifndef DEBUG
+	/* Only allow CoW fork queries if we're debugging. */
+	if (iflags & BMV_IF_COWFORK)
+		return -EINVAL;
+#endif
+	if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
+		return -EINVAL;
+
+	if (iflags & BMV_IF_ATTRFORK)
+		whichfork = XFS_ATTR_FORK;
+	else if (iflags & BMV_IF_COWFORK)
+		whichfork = XFS_COW_FORK;
+	else
+		whichfork = XFS_DATA_FORK;
+
+	switch (whichfork) {
+	case XFS_ATTR_FORK:
 		if (XFS_IFORK_Q(ip)) {
 			if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
 			    ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
@@ -480,7 +577,20 @@ xfs_getbmap(
 
 		prealloced = 0;
 		fixlen = 1LL << 32;
-	} else {
+		break;
+	case XFS_COW_FORK:
+		if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS)
+			return -EINVAL;
+
+		if (xfs_get_cowextsz_hint(ip)) {
+			prealloced = 1;
+			fixlen = mp->m_super->s_maxbytes;
+		} else {
+			prealloced = 0;
+			fixlen = XFS_ISIZE(ip);
+		}
+		break;
+	default:
 		if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 		    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
 		    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
@@ -494,6 +604,7 @@ xfs_getbmap(
 			prealloced = 0;
 			fixlen = XFS_ISIZE(ip);
 		}
+		break;
 	}
 
 	if (bmv->bmv_length == -1) {
@@ -520,7 +631,8 @@ xfs_getbmap(
 		return -ENOMEM;
 
 	xfs_ilock(ip, XFS_IOLOCK_SHARED);
-	if (whichfork == XFS_DATA_FORK) {
+	switch (whichfork) {
+	case XFS_DATA_FORK:
 		if (!(iflags & BMV_IF_DELALLOC) &&
 		    (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
 			error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
@@ -538,8 +650,14 @@ xfs_getbmap(
 		}
 
 		lock = xfs_ilock_data_map_shared(ip);
-	} else {
+		break;
+	case XFS_COW_FORK:
+		lock = XFS_ILOCK_SHARED;
+		xfs_ilock(ip, lock);
+		break;
+	case XFS_ATTR_FORK:
 		lock = xfs_ilock_attr_map_shared(ip);
+		break;
 	}
 
 	/*
@@ -581,7 +699,8 @@ xfs_getbmap(
 			goto out_free_map;
 		ASSERT(nmap <= subnex);
 
-		for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
+		for (i = 0; i < nmap && nexleft && bmv->bmv_length &&
+				cur_ext < bmv->bmv_count; i++) {
 			out[cur_ext].bmv_oflags = 0;
 			if (map[i].br_state == XFS_EXT_UNWRITTEN)
 				out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
@@ -614,9 +733,16 @@ xfs_getbmap(
 				goto out_free_map;
 			}
 
-			if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
-					prealloced, bmvend,
-					map[i].br_startblock))
+			/* Is this a shared block? */
+			error = xfs_getbmap_adjust_shared(ip, whichfork,
+					&map[i], &out[cur_ext], &inject_map);
+			if (error)
+				goto out_free_map;
+
+			if (!xfs_getbmapx_fix_eof_hole(ip, whichfork,
+					&out[cur_ext], prealloced, bmvend,
+					map[i].br_startblock,
+					inject_map.br_startblock != NULLFSBLOCK))
 				goto out_free_map;
 
 			bmv->bmv_offset =
@@ -636,11 +762,16 @@ xfs_getbmap(
 				continue;
 			}
 
-			nexleft--;
+			if (inject_map.br_startblock != NULLFSBLOCK) {
+				map[i] = inject_map;
+				i--;
+			} else
+				nexleft--;
 			bmv->bmv_entries++;
 			cur_ext++;
 		}
-	} while (nmap && nexleft && bmv->bmv_length);
+	} while (nmap && nexleft && bmv->bmv_length &&
+		 cur_ext < bmv->bmv_count);
 
  out_free_map:
 	kmem_free(map);
@@ -1433,8 +1564,8 @@ xfs_insert_file_space(
  */
 static int
 xfs_swap_extents_check_format(
-	xfs_inode_t	*ip,	/* target inode */
-	xfs_inode_t	*tip)	/* tmp inode */
+	struct xfs_inode	*ip,	/* target inode */
+	struct xfs_inode	*tip)	/* tmp inode */
 {
 
 	/* Should never get a local format */
@@ -1449,6 +1580,13 @@ xfs_swap_extents_check_format(
 	if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 		return -EINVAL;
 
+	/*
+	 * If we have to use the (expensive) rmap swap method, we can
+	 * handle any number of extents and any format.
+	 */
+	if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
+		return 0;
+
 	/*
 	 * if the target inode is in extent form and the temp inode is in btree
 	 * form then we will end up with the target inode in the wrong format
@@ -1518,125 +1656,161 @@ xfs_swap_extent_flush(
 	return 0;
 }
 
-int
-xfs_swap_extents(
-	xfs_inode_t	*ip,	/* target inode */
-	xfs_inode_t	*tip,	/* tmp inode */
-	xfs_swapext_t	*sxp)
+/*
+ * Move extents from one file to another, when rmap is enabled.
+ */
+STATIC int
+xfs_swap_extent_rmap(
+	struct xfs_trans		**tpp,
+	struct xfs_inode		*ip,
+	struct xfs_inode		*tip)
 {
-	xfs_mount_t	*mp = ip->i_mount;
-	xfs_trans_t	*tp;
-	xfs_bstat_t	*sbp = &sxp->sx_stat;
-	xfs_ifork_t	*tempifp, *ifp, *tifp;
-	int		src_log_flags, target_log_flags;
-	int		error = 0;
-	int		aforkblks = 0;
-	int		taforkblks = 0;
-	__uint64_t	tmp;
-	int		lock_flags;
-
-	/* XXX: we can't do this with rmap, will fix later */
-	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
-		return -EOPNOTSUPP;
-
-	tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
-	if (!tempifp) {
-		error = -ENOMEM;
-		goto out;
-	}
+	struct xfs_bmbt_irec		irec;
+	struct xfs_bmbt_irec		uirec;
+	struct xfs_bmbt_irec		tirec;
+	xfs_fileoff_t			offset_fsb;
+	xfs_fileoff_t			end_fsb;
+	xfs_filblks_t			count_fsb;
+	xfs_fsblock_t			firstfsb;
+	struct xfs_defer_ops		dfops;
+	int				error;
+	xfs_filblks_t			ilen;
+	xfs_filblks_t			rlen;
+	int				nimaps;
+	__uint64_t			tip_flags2;
 
 	/*
-	 * Lock the inodes against other IO, page faults and truncate to
-	 * begin with.  Then we can ensure the inodes are flushed and have no
-	 * page cache safely. Once we have done this we can take the ilocks and
-	 * do the rest of the checks.
+	 * If the source file has shared blocks, we must flag the donor
+	 * file as having shared blocks so that we get the shared-block
+	 * rmap functions when we go to fix up the rmaps.  The flags
+	 * will be switch for reals later.
 	 */
-	lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
-	xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
-	xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
-
-	/* Verify that both files have the same format */
-	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
-		error = -EINVAL;
-		goto out_unlock;
-	}
+	tip_flags2 = tip->i_d.di_flags2;
+	if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
+		tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
+
+	offset_fsb = 0;
+	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
+	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
+
+	while (count_fsb) {
+		/* Read extent from the donor file */
+		nimaps = 1;
+		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
+				&nimaps, 0);
+		if (error)
+			goto out;
+		ASSERT(nimaps == 1);
+		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
+
+		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
+		ilen = tirec.br_blockcount;
+
+		/* Unmap the old blocks in the source file. */
+		while (tirec.br_blockcount) {
+			xfs_defer_init(&dfops, &firstfsb);
+			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
+
+			/* Read extent from the source file */
+			nimaps = 1;
+			error = xfs_bmapi_read(ip, tirec.br_startoff,
+					tirec.br_blockcount, &irec,
+					&nimaps, 0);
+			if (error)
+				goto out_defer;
+			ASSERT(nimaps == 1);
+			ASSERT(tirec.br_startoff == irec.br_startoff);
+			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
+
+			/* Trim the extent. */
+			uirec = tirec;
+			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
+					tirec.br_blockcount,
+					irec.br_blockcount);
+			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
+
+			/* Remove the mapping from the donor file. */
+			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
+					tip, &uirec);
+			if (error)
+				goto out_defer;
 
-	/* Verify both files are either real-time or non-realtime */
-	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
-		error = -EINVAL;
-		goto out_unlock;
-	}
+			/* Remove the mapping from the source file. */
+			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
+					ip, &irec);
+			if (error)
+				goto out_defer;
 
-	error = xfs_swap_extent_flush(ip);
-	if (error)
-		goto out_unlock;
-	error = xfs_swap_extent_flush(tip);
-	if (error)
-		goto out_unlock;
+			/* Map the donor file's blocks into the source file. */
+			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
+					ip, &uirec);
+			if (error)
+				goto out_defer;
 
-	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
-	if (error)
-		goto out_unlock;
+			/* Map the source file's blocks into the donor file. */
+			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
+					tip, &irec);
+			if (error)
+				goto out_defer;
 
-	/*
-	 * Lock and join the inodes to the tansaction so that transaction commit
-	 * or cancel will unlock the inodes from this point onwards.
-	 */
-	xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
-	lock_flags |= XFS_ILOCK_EXCL;
-	xfs_trans_ijoin(tp, ip, lock_flags);
-	xfs_trans_ijoin(tp, tip, lock_flags);
+			error = xfs_defer_finish(tpp, &dfops, ip);
+			if (error)
+				goto out_defer;
 
+			tirec.br_startoff += rlen;
+			if (tirec.br_startblock != HOLESTARTBLOCK &&
+			    tirec.br_startblock != DELAYSTARTBLOCK)
+				tirec.br_startblock += rlen;
+			tirec.br_blockcount -= rlen;
+		}
 
-	/* Verify all data are being swapped */
-	if (sxp->sx_offset != 0 ||
-	    sxp->sx_length != ip->i_d.di_size ||
-	    sxp->sx_length != tip->i_d.di_size) {
-		error = -EFAULT;
-		goto out_trans_cancel;
+		/* Roll on... */
+		count_fsb -= ilen;
+		offset_fsb += ilen;
 	}
 
-	trace_xfs_swap_extent_before(ip, 0);
-	trace_xfs_swap_extent_before(tip, 1);
+	tip->i_d.di_flags2 = tip_flags2;
+	return 0;
 
-	/* check inode formats now that data is flushed */
-	error = xfs_swap_extents_check_format(ip, tip);
-	if (error) {
-		xfs_notice(mp,
-		    "%s: inode 0x%llx format is incompatible for exchanging.",
-				__func__, ip->i_ino);
-		goto out_trans_cancel;
-	}
+out_defer:
+	xfs_defer_cancel(&dfops);
+out:
+	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
+	tip->i_d.di_flags2 = tip_flags2;
+	return error;
+}
+
+/* Swap the extents of two files by swapping data forks. */
+STATIC int
+xfs_swap_extent_forks(
+	struct xfs_trans	*tp,
+	struct xfs_inode	*ip,
+	struct xfs_inode	*tip,
+	int			*src_log_flags,
+	int			*target_log_flags)
+{
+	struct xfs_ifork	tempifp, *ifp, *tifp;
+	int			aforkblks = 0;
+	int			taforkblks = 0;
+	__uint64_t		tmp;
+	int			error;
 
-	/*
-	 * Compare the current change & modify times with that
-	 * passed in.  If they differ, we abort this swap.
-	 * This is the mechanism used to ensure the calling
-	 * process that the file was not changed out from
-	 * under it.
-	 */
-	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
-	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
-	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
-	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
-		error = -EBUSY;
-		goto out_trans_cancel;
-	}
 	/*
 	 * Count the number of extended attribute blocks
 	 */
 	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
-		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
+		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK,
+				&aforkblks);
 		if (error)
-			goto out_trans_cancel;
+			return error;
 	}
 	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
-			&taforkblks);
+				&taforkblks);
 		if (error)
-			goto out_trans_cancel;
+			return error;
 	}
 
 	/*
@@ -1645,31 +1819,23 @@ xfs_swap_extents(
 	 * buffers, and so the validation done on read will expect the owner
 	 * field to be correctly set. Once we change the owners, we can swap the
 	 * inode forks.
-	 *
-	 * Note the trickiness in setting the log flags - we set the owner log
-	 * flag on the opposite inode (i.e. the inode we are setting the new
-	 * owner to be) because once we swap the forks and log that, log
-	 * recovery is going to see the fork as owned by the swapped inode,
-	 * not the pre-swapped inodes.
 	 */
-	src_log_flags = XFS_ILOG_CORE;
-	target_log_flags = XFS_ILOG_CORE;
 	if (ip->i_d.di_version == 3 &&
 	    ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
-		target_log_flags |= XFS_ILOG_DOWNER;
+		(*target_log_flags) |= XFS_ILOG_DOWNER;
 		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
 					      tip->i_ino, NULL);
 		if (error)
-			goto out_trans_cancel;
+			return error;
 	}
 
 	if (tip->i_d.di_version == 3 &&
 	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
-		src_log_flags |= XFS_ILOG_DOWNER;
+		(*src_log_flags) |= XFS_ILOG_DOWNER;
 		error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
 					      ip->i_ino, NULL);
 		if (error)
-			goto out_trans_cancel;
+			return error;
 	}
 
 	/*
@@ -1677,9 +1843,9 @@ xfs_swap_extents(
 	 */
 	ifp = &ip->i_df;
 	tifp = &tip->i_df;
-	*tempifp = *ifp;	/* struct copy */
+	tempifp = *ifp;		/* struct copy */
 	*ifp = *tifp;		/* struct copy */
-	*tifp = *tempifp;	/* struct copy */
+	*tifp = tempifp;	/* struct copy */
 
 	/*
 	 * Fix the on-disk inode values
@@ -1719,12 +1885,12 @@ xfs_swap_extents(
 			ifp->if_u1.if_extents =
 				ifp->if_u2.if_inline_ext;
 		}
-		src_log_flags |= XFS_ILOG_DEXT;
+		(*src_log_flags) |= XFS_ILOG_DEXT;
 		break;
 	case XFS_DINODE_FMT_BTREE:
 		ASSERT(ip->i_d.di_version < 3 ||
-		       (src_log_flags & XFS_ILOG_DOWNER));
-		src_log_flags |= XFS_ILOG_DBROOT;
+		       (*src_log_flags & XFS_ILOG_DOWNER));
+		(*src_log_flags) |= XFS_ILOG_DBROOT;
 		break;
 	}
 
@@ -1738,15 +1904,166 @@ xfs_swap_extents(
 			tifp->if_u1.if_extents =
 				tifp->if_u2.if_inline_ext;
 		}
-		target_log_flags |= XFS_ILOG_DEXT;
+		(*target_log_flags) |= XFS_ILOG_DEXT;
 		break;
 	case XFS_DINODE_FMT_BTREE:
-		target_log_flags |= XFS_ILOG_DBROOT;
+		(*target_log_flags) |= XFS_ILOG_DBROOT;
 		ASSERT(tip->i_d.di_version < 3 ||
-		       (target_log_flags & XFS_ILOG_DOWNER));
+		       (*target_log_flags & XFS_ILOG_DOWNER));
 		break;
 	}
 
+	return 0;
+}
+
+int
+xfs_swap_extents(
+	struct xfs_inode	*ip,	/* target inode */
+	struct xfs_inode	*tip,	/* tmp inode */
+	struct xfs_swapext	*sxp)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	struct xfs_bstat	*sbp = &sxp->sx_stat;
+	int			src_log_flags, target_log_flags;
+	int			error = 0;
+	int			lock_flags;
+	struct xfs_ifork	*cowfp;
+	__uint64_t		f;
+	int			resblks;
+
+	/*
+	 * Lock the inodes against other IO, page faults and truncate to
+	 * begin with.  Then we can ensure the inodes are flushed and have no
+	 * page cache safely. Once we have done this we can take the ilocks and
+	 * do the rest of the checks.
+	 */
+	lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
+	xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
+	xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
+
+	/* Verify that both files have the same format */
+	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
+		error = -EINVAL;
+		goto out_unlock;
+	}
+
+	/* Verify both files are either real-time or non-realtime */
+	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
+		error = -EINVAL;
+		goto out_unlock;
+	}
+
+	error = xfs_swap_extent_flush(ip);
+	if (error)
+		goto out_unlock;
+	error = xfs_swap_extent_flush(tip);
+	if (error)
+		goto out_unlock;
+
+	/*
+	 * Extent "swapping" with rmap requires a permanent reservation and
+	 * a block reservation because it's really just a remap operation
+	 * performed with log redo items!
+	 */
+	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
+		/*
+		 * Conceptually this shouldn't affect the shape of either
+		 * bmbt, but since we atomically move extents one by one,
+		 * we reserve enough space to rebuild both trees.
+		 */
+		resblks = XFS_SWAP_RMAP_SPACE_RES(mp,
+				XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK),
+				XFS_DATA_FORK) +
+			  XFS_SWAP_RMAP_SPACE_RES(mp,
+				XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK),
+				XFS_DATA_FORK);
+		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
+				0, 0, &tp);
+	} else
+		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0,
+				0, 0, &tp);
+	if (error)
+		goto out_unlock;
+
+	/*
+	 * Lock and join the inodes to the tansaction so that transaction commit
+	 * or cancel will unlock the inodes from this point onwards.
+	 */
+	xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
+	lock_flags |= XFS_ILOCK_EXCL;
+	xfs_trans_ijoin(tp, ip, 0);
+	xfs_trans_ijoin(tp, tip, 0);
+
+
+	/* Verify all data are being swapped */
+	if (sxp->sx_offset != 0 ||
+	    sxp->sx_length != ip->i_d.di_size ||
+	    sxp->sx_length != tip->i_d.di_size) {
+		error = -EFAULT;
+		goto out_trans_cancel;
+	}
+
+	trace_xfs_swap_extent_before(ip, 0);
+	trace_xfs_swap_extent_before(tip, 1);
+
+	/* check inode formats now that data is flushed */
+	error = xfs_swap_extents_check_format(ip, tip);
+	if (error) {
+		xfs_notice(mp,
+		    "%s: inode 0x%llx format is incompatible for exchanging.",
+				__func__, ip->i_ino);
+		goto out_trans_cancel;
+	}
+
+	/*
+	 * Compare the current change & modify times with that
+	 * passed in.  If they differ, we abort this swap.
+	 * This is the mechanism used to ensure the calling
+	 * process that the file was not changed out from
+	 * under it.
+	 */
+	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
+	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
+	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
+	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
+		error = -EBUSY;
+		goto out_trans_cancel;
+	}
+
+	/*
+	 * Note the trickiness in setting the log flags - we set the owner log
+	 * flag on the opposite inode (i.e. the inode we are setting the new
+	 * owner to be) because once we swap the forks and log that, log
+	 * recovery is going to see the fork as owned by the swapped inode,
+	 * not the pre-swapped inodes.
+	 */
+	src_log_flags = XFS_ILOG_CORE;
+	target_log_flags = XFS_ILOG_CORE;
+
+	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
+		error = xfs_swap_extent_rmap(&tp, ip, tip);
+	else
+		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
+				&target_log_flags);
+	if (error)
+		goto out_trans_cancel;
+
+	/* Do we have to swap reflink flags? */
+	if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
+	    (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
+		f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
+		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+		ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
+		tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+		tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
+		cowfp = ip->i_cowfp;
+		ip->i_cowfp = tip->i_cowfp;
+		tip->i_cowfp = cowfp;
+		xfs_inode_set_cowblocks_tag(ip);
+		xfs_inode_set_cowblocks_tag(tip);
+	}
+
 	xfs_trans_log_inode(tp, ip,  src_log_flags);
 	xfs_trans_log_inode(tp, tip, target_log_flags);
 
@@ -1761,16 +2078,16 @@ xfs_swap_extents(
 
 	trace_xfs_swap_extent_after(ip, 0);
 	trace_xfs_swap_extent_after(tip, 1);
-out:
-	kmem_free(tempifp);
-	return error;
 
-out_unlock:
 	xfs_iunlock(ip, lock_flags);
 	xfs_iunlock(tip, lock_flags);
-	goto out;
+	return error;
 
 out_trans_cancel:
 	xfs_trans_cancel(tp);
-	goto out;
+
+out_unlock:
+	xfs_iunlock(ip, lock_flags);
+	xfs_iunlock(tip, lock_flags);
+	return error;
 }

fs/xfs/xfs_dir2_readdir.c

@@ -84,7 +84,8 @@ xfs_dir2_sf_getdents(
 
 	sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
 
-	ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
+	if (dp->i_d.di_size < xfs_dir2_sf_hdr_size(sfp->i8count))
+		return -EFSCORRUPTED;
 
 	/*
 	 * If the block number in the offset is out of range, we're done.

fs/xfs/xfs_error.h

@@ -92,7 +92,11 @@ extern void xfs_verifier_error(struct xfs_buf *bp);
 #define XFS_ERRTAG_BMAPIFORMAT				21
 #define XFS_ERRTAG_FREE_EXTENT				22
 #define XFS_ERRTAG_RMAP_FINISH_ONE			23
-#define XFS_ERRTAG_MAX					24
+#define XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE		24
+#define XFS_ERRTAG_REFCOUNT_FINISH_ONE			25
+#define XFS_ERRTAG_BMAP_FINISH_ONE			26
+#define XFS_ERRTAG_AG_RESV_CRITICAL			27
+#define XFS_ERRTAG_MAX					28
 
 /*
  * Random factors for above tags, 1 means always, 2 means 1/2 time, etc.
@@ -121,6 +125,10 @@ extern void xfs_verifier_error(struct xfs_buf *bp);
 #define	XFS_RANDOM_BMAPIFORMAT				XFS_RANDOM_DEFAULT
 #define XFS_RANDOM_FREE_EXTENT				1
 #define XFS_RANDOM_RMAP_FINISH_ONE			1
+#define XFS_RANDOM_REFCOUNT_CONTINUE_UPDATE		1
+#define XFS_RANDOM_REFCOUNT_FINISH_ONE			1
+#define XFS_RANDOM_BMAP_FINISH_ONE			1
+#define XFS_RANDOM_AG_RESV_CRITICAL			4
 
 #ifdef DEBUG
 extern int xfs_error_test_active;

fs/xfs/xfs_file.c

@@ -38,6 +38,7 @@
 #include "xfs_icache.h"
 #include "xfs_pnfs.h"
 #include "xfs_iomap.h"
+#include "xfs_reflink.h"
 
 #include <linux/dcache.h>
 #include <linux/falloc.h>
@@ -634,6 +635,13 @@ xfs_file_dio_aio_write(
 
 	trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
 
+	/* If this is a block-aligned directio CoW, remap immediately. */
+	if (xfs_is_reflink_inode(ip) && !unaligned_io) {
+		ret = xfs_reflink_allocate_cow_range(ip, iocb->ki_pos, count);
+		if (ret)
+			goto out;
+	}
+
 	data = *from;
 	ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
 			xfs_get_blocks_direct, xfs_end_io_direct_write,
@@ -735,6 +743,9 @@ write_retry:
 		enospc = xfs_inode_free_quota_eofblocks(ip);
 		if (enospc)
 			goto write_retry;
+		enospc = xfs_inode_free_quota_cowblocks(ip);
+		if (enospc)
+			goto write_retry;
 	} else if (ret == -ENOSPC && !enospc) {
 		struct xfs_eofblocks eofb = {0};
 
@@ -774,10 +785,20 @@ xfs_file_write_iter(
 
 	if (IS_DAX(inode))
 		ret = xfs_file_dax_write(iocb, from);
-	else if (iocb->ki_flags & IOCB_DIRECT)
+	else if (iocb->ki_flags & IOCB_DIRECT) {
+		/*
+		 * Allow a directio write to fall back to a buffered
+		 * write *only* in the case that we're doing a reflink
+		 * CoW.  In all other directio scenarios we do not
+		 * allow an operation to fall back to buffered mode.
+		 */
 		ret = xfs_file_dio_aio_write(iocb, from);
-	else
+		if (ret == -EREMCHG)
+			goto buffered;
+	} else {
+buffered:
 		ret = xfs_file_buffered_aio_write(iocb, from);
+	}
 
 	if (ret > 0) {
 		XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
@@ -791,7 +812,7 @@ xfs_file_write_iter(
 #define	XFS_FALLOC_FL_SUPPORTED						\
 		(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |		\
 		 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |	\
-		 FALLOC_FL_INSERT_RANGE)
+		 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
 
 STATIC long
 xfs_file_fallocate(
@@ -881,9 +902,15 @@ xfs_file_fallocate(
 
 		if (mode & FALLOC_FL_ZERO_RANGE)
 			error = xfs_zero_file_space(ip, offset, len);
-		else
+		else {
+			if (mode & FALLOC_FL_UNSHARE_RANGE) {
+				error = xfs_reflink_unshare(ip, offset, len);
+				if (error)
+					goto out_unlock;
+			}
 			error = xfs_alloc_file_space(ip, offset, len,
 						     XFS_BMAPI_PREALLOC);
+		}
 		if (error)
 			goto out_unlock;
 	}
@@ -920,6 +947,189 @@ out_unlock:
 	return error;
 }
 
+/*
+ * Flush all file writes out to disk.
+ */
+static int
+xfs_file_wait_for_io(
+	struct inode	*inode,
+	loff_t		offset,
+	size_t		len)
+{
+	loff_t		rounding;
+	loff_t		ioffset;
+	loff_t		iendoffset;
+	loff_t		bs;
+	int		ret;
+
+	bs = inode->i_sb->s_blocksize;
+	inode_dio_wait(inode);
+
+	rounding = max_t(xfs_off_t, bs, PAGE_SIZE);
+	ioffset = round_down(offset, rounding);
+	iendoffset = round_up(offset + len, rounding) - 1;
+	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
+					   iendoffset);
+	return ret;
+}
+
+/* Hook up to the VFS reflink function */
+STATIC int
+xfs_file_share_range(
+	struct file	*file_in,
+	loff_t		pos_in,
+	struct file	*file_out,
+	loff_t		pos_out,
+	u64		len,
+	bool		is_dedupe)
+{
+	struct inode	*inode_in;
+	struct inode	*inode_out;
+	ssize_t		ret;
+	loff_t		bs;
+	loff_t		isize;
+	int		same_inode;
+	loff_t		blen;
+	unsigned int	flags = 0;
+
+	inode_in = file_inode(file_in);
+	inode_out = file_inode(file_out);
+	bs = inode_out->i_sb->s_blocksize;
+
+	/* Don't touch certain kinds of inodes */
+	if (IS_IMMUTABLE(inode_out))
+		return -EPERM;
+	if (IS_SWAPFILE(inode_in) ||
+	    IS_SWAPFILE(inode_out))
+		return -ETXTBSY;
+
+	/* Reflink only works within this filesystem. */
+	if (inode_in->i_sb != inode_out->i_sb)
+		return -EXDEV;
+	same_inode = (inode_in->i_ino == inode_out->i_ino);
+
+	/* Don't reflink dirs, pipes, sockets... */
+	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
+		return -EISDIR;
+	if (S_ISFIFO(inode_in->i_mode) || S_ISFIFO(inode_out->i_mode))
+		return -EINVAL;
+	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
+		return -EINVAL;
+
+	/* Don't share DAX file data for now. */
+	if (IS_DAX(inode_in) || IS_DAX(inode_out))
+		return -EINVAL;
+
+	/* Are we going all the way to the end? */
+	isize = i_size_read(inode_in);
+	if (isize == 0)
+		return 0;
+	if (len == 0)
+		len = isize - pos_in;
+
+	/* Ensure offsets don't wrap and the input is inside i_size */
+	if (pos_in + len < pos_in || pos_out + len < pos_out ||
+	    pos_in + len > isize)
+		return -EINVAL;
+
+	/* Don't allow dedupe past EOF in the dest file */
+	if (is_dedupe) {
+		loff_t	disize;
+
+		disize = i_size_read(inode_out);
+		if (pos_out >= disize || pos_out + len > disize)
+			return -EINVAL;
+	}
+
+	/* If we're linking to EOF, continue to the block boundary. */
+	if (pos_in + len == isize)
+		blen = ALIGN(isize, bs) - pos_in;
+	else
+		blen = len;
+
+	/* Only reflink if we're aligned to block boundaries */
+	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
+	    !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
+		return -EINVAL;
+
+	/* Don't allow overlapped reflink within the same file */
+	if (same_inode && pos_out + blen > pos_in && pos_out < pos_in + blen)
+		return -EINVAL;
+
+	/* Wait for the completion of any pending IOs on srcfile */
+	ret = xfs_file_wait_for_io(inode_in, pos_in, len);
+	if (ret)
+		goto out;
+	ret = xfs_file_wait_for_io(inode_out, pos_out, len);
+	if (ret)
+		goto out;
+
+	if (is_dedupe)
+		flags |= XFS_REFLINK_DEDUPE;
+	ret = xfs_reflink_remap_range(XFS_I(inode_in), pos_in, XFS_I(inode_out),
+			pos_out, len, flags);
+	if (ret < 0)
+		goto out;
+
+out:
+	return ret;
+}
+
+STATIC ssize_t
+xfs_file_copy_range(
+	struct file	*file_in,
+	loff_t		pos_in,
+	struct file	*file_out,
+	loff_t		pos_out,
+	size_t		len,
+	unsigned int	flags)
+{
+	int		error;
+
+	error = xfs_file_share_range(file_in, pos_in, file_out, pos_out,
+				     len, false);
+	if (error)
+		return error;
+	return len;
+}
+
+STATIC int
+xfs_file_clone_range(
+	struct file	*file_in,
+	loff_t		pos_in,
+	struct file	*file_out,
+	loff_t		pos_out,
+	u64		len)
+{
+	return xfs_file_share_range(file_in, pos_in, file_out, pos_out,
+				     len, false);
+}
+
+#define XFS_MAX_DEDUPE_LEN	(16 * 1024 * 1024)
+STATIC ssize_t
+xfs_file_dedupe_range(
+	struct file	*src_file,
+	u64		loff,
+	u64		len,
+	struct file	*dst_file,
+	u64		dst_loff)
+{
+	int		error;
+
+	/*
+	 * Limit the total length we will dedupe for each operation.
+	 * This is intended to bound the total time spent in this
+	 * ioctl to something sane.
+	 */
+	if (len > XFS_MAX_DEDUPE_LEN)
+		len = XFS_MAX_DEDUPE_LEN;
+
+	error = xfs_file_share_range(src_file, loff, dst_file, dst_loff,
+				     len, true);
+	if (error)
+		return error;
+	return len;
+}
 
 STATIC int
 xfs_file_open(
@@ -1581,6 +1791,9 @@ const struct file_operations xfs_file_operations = {
 	.fsync		= xfs_file_fsync,
 	.get_unmapped_area = thp_get_unmapped_area,
 	.fallocate	= xfs_file_fallocate,
+	.copy_file_range = xfs_file_copy_range,
+	.clone_file_range = xfs_file_clone_range,
+	.dedupe_file_range = xfs_file_dedupe_range,
 };
 
 const struct file_operations xfs_dir_file_operations = {

fs/xfs/xfs_fsops.c

@@ -43,6 +43,7 @@
 #include "xfs_log.h"
 #include "xfs_filestream.h"
 #include "xfs_rmap.h"
+#include "xfs_ag_resv.h"
 
 /*
  * File system operations
@@ -108,7 +109,9 @@ xfs_fs_geometry(
 			(xfs_sb_version_hassparseinodes(&mp->m_sb) ?
 				XFS_FSOP_GEOM_FLAGS_SPINODES : 0) |
 			(xfs_sb_version_hasrmapbt(&mp->m_sb) ?
-				XFS_FSOP_GEOM_FLAGS_RMAPBT : 0);
+				XFS_FSOP_GEOM_FLAGS_RMAPBT : 0) |
+			(xfs_sb_version_hasreflink(&mp->m_sb) ?
+				XFS_FSOP_GEOM_FLAGS_REFLINK : 0);
 		geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 				mp->m_sb.sb_logsectsize : BBSIZE;
 		geo->rtsectsize = mp->m_sb.sb_blocksize;
@@ -259,6 +262,12 @@ xfs_growfs_data_private(
 		agf->agf_longest = cpu_to_be32(tmpsize);
 		if (xfs_sb_version_hascrc(&mp->m_sb))
 			uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
+		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
+			agf->agf_refcount_root = cpu_to_be32(
+					xfs_refc_block(mp));
+			agf->agf_refcount_level = cpu_to_be32(1);
+			agf->agf_refcount_blocks = cpu_to_be32(1);
+		}
 
 		error = xfs_bwrite(bp);
 		xfs_buf_relse(bp);
@@ -450,6 +459,17 @@ xfs_growfs_data_private(
 			rrec->rm_offset = 0;
 			be16_add_cpu(&block->bb_numrecs, 1);
 
+			/* account for refc btree root */
+			if (xfs_sb_version_hasreflink(&mp->m_sb)) {
+				rrec = XFS_RMAP_REC_ADDR(block, 5);
+				rrec->rm_startblock = cpu_to_be32(
+						xfs_refc_block(mp));
+				rrec->rm_blockcount = cpu_to_be32(1);
+				rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
+				rrec->rm_offset = 0;
+				be16_add_cpu(&block->bb_numrecs, 1);
+			}
+
 			error = xfs_bwrite(bp);
 			xfs_buf_relse(bp);
 			if (error)
@@ -507,6 +527,28 @@ xfs_growfs_data_private(
 				goto error0;
 		}
 
+		/*
+		 * refcount btree root block
+		 */
+		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
+			bp = xfs_growfs_get_hdr_buf(mp,
+				XFS_AGB_TO_DADDR(mp, agno, xfs_refc_block(mp)),
+				BTOBB(mp->m_sb.sb_blocksize), 0,
+				&xfs_refcountbt_buf_ops);
+			if (!bp) {
+				error = -ENOMEM;
+				goto error0;
+			}
+
+			xfs_btree_init_block(mp, bp, XFS_REFC_CRC_MAGIC,
+					     0, 0, agno,
+					     XFS_BTREE_CRC_BLOCKS);
+
+			error = xfs_bwrite(bp);
+			xfs_buf_relse(bp);
+			if (error)
+				goto error0;
+		}
 	}
 	xfs_trans_agblocks_delta(tp, nfree);
 	/*
@@ -589,6 +631,11 @@ xfs_growfs_data_private(
 	xfs_set_low_space_thresholds(mp);
 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
 
+	/* Reserve AG metadata blocks. */
+	error = xfs_fs_reserve_ag_blocks(mp);
+	if (error && error != -ENOSPC)
+		goto out;
+
 	/* update secondary superblocks. */
 	for (agno = 1; agno < nagcount; agno++) {
 		error = 0;
@@ -639,6 +686,8 @@ xfs_growfs_data_private(
 			continue;
 		}
 	}
+
+ out:
 	return saved_error ? saved_error : error;
 
  error0:
@@ -948,3 +997,59 @@ xfs_do_force_shutdown(
 	"Please umount the filesystem and rectify the problem(s)");
 	}
 }
+
+/*
+ * Reserve free space for per-AG metadata.
+ */
+int
+xfs_fs_reserve_ag_blocks(
+	struct xfs_mount	*mp)
+{
+	xfs_agnumber_t		agno;
+	struct xfs_perag	*pag;
+	int			error = 0;
+	int			err2;
+
+	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
+		pag = xfs_perag_get(mp, agno);
+		err2 = xfs_ag_resv_init(pag);
+		xfs_perag_put(pag);
+		if (err2 && !error)
+			error = err2;
+	}
+
+	if (error && error != -ENOSPC) {
+		xfs_warn(mp,
+	"Error %d reserving per-AG metadata reserve pool.", error);
+		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+	}
+
+	return error;
+}
+
+/*
+ * Free space reserved for per-AG metadata.
+ */
+int
+xfs_fs_unreserve_ag_blocks(
+	struct xfs_mount	*mp)
+{
+	xfs_agnumber_t		agno;
+	struct xfs_perag	*pag;
+	int			error = 0;
+	int			err2;
+
+	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
+		pag = xfs_perag_get(mp, agno);
+		err2 = xfs_ag_resv_free(pag);
+		xfs_perag_put(pag);
+		if (err2 && !error)
+			error = err2;
+	}
+
+	if (error)
+		xfs_warn(mp,
+	"Error %d freeing per-AG metadata reserve pool.", error);
+
+	return error;
+}

fs/xfs/xfs_fsops.h

@@ -26,4 +26,7 @@ extern int xfs_reserve_blocks(xfs_mount_t *mp, __uint64_t *inval,
 				xfs_fsop_resblks_t *outval);
 extern int xfs_fs_goingdown(xfs_mount_t *mp, __uint32_t inflags);
 
+extern int xfs_fs_reserve_ag_blocks(struct xfs_mount *mp);
+extern int xfs_fs_unreserve_ag_blocks(struct xfs_mount *mp);
+
 #endif	/* __XFS_FSOPS_H__ */

fs/xfs/xfs_globals.c

@@ -21,8 +21,8 @@
 /*
  * Tunable XFS parameters.  xfs_params is required even when CONFIG_SYSCTL=n,
  * other XFS code uses these values.  Times are measured in centisecs (i.e.
- * 100ths of a second) with the exception of eofb_timer, which is measured in
- * seconds.
+ * 100ths of a second) with the exception of eofb_timer and cowb_timer, which
+ * are measured in seconds.
  */
 xfs_param_t xfs_params = {
 			  /*	MIN		DFLT		MAX	*/
@@ -42,6 +42,7 @@ xfs_param_t xfs_params = {
 	.inherit_nodfrg	= {	0,		1,		1	},
 	.fstrm_timer	= {	1,		30*100,		3600*100},
 	.eofb_timer	= {	1,		300,		3600*24},
+	.cowb_timer	= {	1,		1800,		3600*24},
 };
 
 struct xfs_globals xfs_globals = {

fs/xfs/xfs_icache.c

@@ -33,6 +33,7 @@
 #include "xfs_bmap_util.h"
 #include "xfs_dquot_item.h"
 #include "xfs_dquot.h"
+#include "xfs_reflink.h"
 
 #include <linux/kthread.h>
 #include <linux/freezer.h>
@@ -76,6 +77,9 @@ xfs_inode_alloc(
 	ip->i_mount = mp;
 	memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
 	ip->i_afp = NULL;
+	ip->i_cowfp = NULL;
+	ip->i_cnextents = 0;
+	ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
 	memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
 	ip->i_flags = 0;
 	ip->i_delayed_blks = 0;
@@ -101,6 +105,8 @@ xfs_inode_free_callback(
 
 	if (ip->i_afp)
 		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
+	if (ip->i_cowfp)
+		xfs_idestroy_fork(ip, XFS_COW_FORK);
 
 	if (ip->i_itemp) {
 		ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL));
@@ -787,6 +793,33 @@ xfs_eofblocks_worker(
 	xfs_queue_eofblocks(mp);
 }
 
+/*
+ * Background scanning to trim preallocated CoW space. This is queued
+ * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
+ * (We'll just piggyback on the post-EOF prealloc space workqueue.)
+ */
+STATIC void
+xfs_queue_cowblocks(
+	struct xfs_mount *mp)
+{
+	rcu_read_lock();
+	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
+		queue_delayed_work(mp->m_eofblocks_workqueue,
+				   &mp->m_cowblocks_work,
+				   msecs_to_jiffies(xfs_cowb_secs * 1000));
+	rcu_read_unlock();
+}
+
+void
+xfs_cowblocks_worker(
+	struct work_struct *work)
+{
+	struct xfs_mount *mp = container_of(to_delayed_work(work),
+				struct xfs_mount, m_cowblocks_work);
+	xfs_icache_free_cowblocks(mp, NULL);
+	xfs_queue_cowblocks(mp);
+}
+
 int
 xfs_inode_ag_iterator(
 	struct xfs_mount	*mp,
@@ -1343,18 +1376,30 @@ xfs_inode_free_eofblocks(
 	return ret;
 }
 
-int
-xfs_icache_free_eofblocks(
+static int
+__xfs_icache_free_eofblocks(
 	struct xfs_mount	*mp,
-	struct xfs_eofblocks	*eofb)
+	struct xfs_eofblocks	*eofb,
+	int			(*execute)(struct xfs_inode *ip, int flags,
+					   void *args),
+	int			tag)
 {
 	int flags = SYNC_TRYLOCK;
 
 	if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
 		flags = SYNC_WAIT;
 
-	return xfs_inode_ag_iterator_tag(mp, xfs_inode_free_eofblocks, flags,
-					 eofb, XFS_ICI_EOFBLOCKS_TAG);
+	return xfs_inode_ag_iterator_tag(mp, execute, flags,
+					 eofb, tag);
+}
+
+int
+xfs_icache_free_eofblocks(
+	struct xfs_mount	*mp,
+	struct xfs_eofblocks	*eofb)
+{
+	return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks,
+			XFS_ICI_EOFBLOCKS_TAG);
 }
 
 /*
@@ -1363,9 +1408,11 @@ xfs_icache_free_eofblocks(
  * failure. We make a best effort by including each quota under low free space
  * conditions (less than 1% free space) in the scan.
  */
-int
-xfs_inode_free_quota_eofblocks(
-	struct xfs_inode *ip)
+static int
+__xfs_inode_free_quota_eofblocks(
+	struct xfs_inode	*ip,
+	int			(*execute)(struct xfs_mount *mp,
+					   struct xfs_eofblocks	*eofb))
 {
 	int scan = 0;
 	struct xfs_eofblocks eofb = {0};
@@ -1401,14 +1448,25 @@ xfs_inode_free_quota_eofblocks(
 	}
 
 	if (scan)
-		xfs_icache_free_eofblocks(ip->i_mount, &eofb);
+		execute(ip->i_mount, &eofb);
 
 	return scan;
 }
 
-void
-xfs_inode_set_eofblocks_tag(
-	xfs_inode_t	*ip)
+int
+xfs_inode_free_quota_eofblocks(
+	struct xfs_inode *ip)
+{
+	return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
+}
+
+static void
+__xfs_inode_set_eofblocks_tag(
+	xfs_inode_t	*ip,
+	void		(*execute)(struct xfs_mount *mp),
+	void		(*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
+				  int error, unsigned long caller_ip),
+	int		tag)
 {
 	struct xfs_mount *mp = ip->i_mount;
 	struct xfs_perag *pag;
@@ -1426,26 +1484,22 @@ xfs_inode_set_eofblocks_tag(
 
 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 	spin_lock(&pag->pag_ici_lock);
-	trace_xfs_inode_set_eofblocks_tag(ip);
 
-	tagged = radix_tree_tagged(&pag->pag_ici_root,
-				   XFS_ICI_EOFBLOCKS_TAG);
+	tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
 	radix_tree_tag_set(&pag->pag_ici_root,
-			   XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
-			   XFS_ICI_EOFBLOCKS_TAG);
+			   XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
 	if (!tagged) {
 		/* propagate the eofblocks tag up into the perag radix tree */
 		spin_lock(&ip->i_mount->m_perag_lock);
 		radix_tree_tag_set(&ip->i_mount->m_perag_tree,
 				   XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
-				   XFS_ICI_EOFBLOCKS_TAG);
+				   tag);
 		spin_unlock(&ip->i_mount->m_perag_lock);
 
 		/* kick off background trimming */
-		xfs_queue_eofblocks(ip->i_mount);
+		execute(ip->i_mount);
 
-		trace_xfs_perag_set_eofblocks(ip->i_mount, pag->pag_agno,
-					      -1, _RET_IP_);
+		set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
 	}
 
 	spin_unlock(&pag->pag_ici_lock);
@@ -1453,8 +1507,21 @@ xfs_inode_set_eofblocks_tag(
 }
 
 void
-xfs_inode_clear_eofblocks_tag(
+xfs_inode_set_eofblocks_tag(
 	xfs_inode_t	*ip)
+{
+	trace_xfs_inode_set_eofblocks_tag(ip);
+	return __xfs_inode_set_eofblocks_tag(ip, xfs_queue_eofblocks,
+			trace_xfs_perag_set_eofblocks,
+			XFS_ICI_EOFBLOCKS_TAG);
+}
+
+static void
+__xfs_inode_clear_eofblocks_tag(
+	xfs_inode_t	*ip,
+	void		(*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
+				    int error, unsigned long caller_ip),
+	int		tag)
 {
 	struct xfs_mount *mp = ip->i_mount;
 	struct xfs_perag *pag;
@@ -1465,23 +1532,141 @@ xfs_inode_clear_eofblocks_tag(
 
 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 	spin_lock(&pag->pag_ici_lock);
-	trace_xfs_inode_clear_eofblocks_tag(ip);
 
 	radix_tree_tag_clear(&pag->pag_ici_root,
-			     XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
-			     XFS_ICI_EOFBLOCKS_TAG);
-	if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_EOFBLOCKS_TAG)) {
+			     XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
+	if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
 		/* clear the eofblocks tag from the perag radix tree */
 		spin_lock(&ip->i_mount->m_perag_lock);
 		radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
 				     XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
-				     XFS_ICI_EOFBLOCKS_TAG);
+				     tag);
 		spin_unlock(&ip->i_mount->m_perag_lock);
-		trace_xfs_perag_clear_eofblocks(ip->i_mount, pag->pag_agno,
-					       -1, _RET_IP_);
+		clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
 	}
 
 	spin_unlock(&pag->pag_ici_lock);
 	xfs_perag_put(pag);
 }
 
+void
+xfs_inode_clear_eofblocks_tag(
+	xfs_inode_t	*ip)
+{
+	trace_xfs_inode_clear_eofblocks_tag(ip);
+	return __xfs_inode_clear_eofblocks_tag(ip,
+			trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
+}
+
+/*
+ * Automatic CoW Reservation Freeing
+ *
+ * These functions automatically garbage collect leftover CoW reservations
+ * that were made on behalf of a cowextsize hint when we start to run out
+ * of quota or when the reservations sit around for too long.  If the file
+ * has dirty pages or is undergoing writeback, its CoW reservations will
+ * be retained.
+ *
+ * The actual garbage collection piggybacks off the same code that runs
+ * the speculative EOF preallocation garbage collector.
+ */
+STATIC int
+xfs_inode_free_cowblocks(
+	struct xfs_inode	*ip,
+	int			flags,
+	void			*args)
+{
+	int ret;
+	struct xfs_eofblocks *eofb = args;
+	bool need_iolock = true;
+	int match;
+
+	ASSERT(!eofb || (eofb && eofb->eof_scan_owner != 0));
+
+	if (!xfs_reflink_has_real_cow_blocks(ip)) {
+		trace_xfs_inode_free_cowblocks_invalid(ip);
+		xfs_inode_clear_cowblocks_tag(ip);
+		return 0;
+	}
+
+	/*
+	 * If the mapping is dirty or under writeback we cannot touch the
+	 * CoW fork.  Leave it alone if we're in the midst of a directio.
+	 */
+	if (mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
+	    mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
+	    atomic_read(&VFS_I(ip)->i_dio_count))
+		return 0;
+
+	if (eofb) {
+		if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
+			match = xfs_inode_match_id_union(ip, eofb);
+		else
+			match = xfs_inode_match_id(ip, eofb);
+		if (!match)
+			return 0;
+
+		/* skip the inode if the file size is too small */
+		if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
+		    XFS_ISIZE(ip) < eofb->eof_min_file_size)
+			return 0;
+
+		/*
+		 * A scan owner implies we already hold the iolock. Skip it in
+		 * xfs_free_eofblocks() to avoid deadlock. This also eliminates
+		 * the possibility of EAGAIN being returned.
+		 */
+		if (eofb->eof_scan_owner == ip->i_ino)
+			need_iolock = false;
+	}
+
+	/* Free the CoW blocks */
+	if (need_iolock) {
+		xfs_ilock(ip, XFS_IOLOCK_EXCL);
+		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
+	}
+
+	ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF);
+
+	if (need_iolock) {
+		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
+		xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+	}
+
+	return ret;
+}
+
+int
+xfs_icache_free_cowblocks(
+	struct xfs_mount	*mp,
+	struct xfs_eofblocks	*eofb)
+{
+	return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks,
+			XFS_ICI_COWBLOCKS_TAG);
+}
+
+int
+xfs_inode_free_quota_cowblocks(
+	struct xfs_inode *ip)
+{
+	return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
+}
+
+void
+xfs_inode_set_cowblocks_tag(
+	xfs_inode_t	*ip)
+{
+	trace_xfs_inode_set_eofblocks_tag(ip);
+	return __xfs_inode_set_eofblocks_tag(ip, xfs_queue_cowblocks,
+			trace_xfs_perag_set_eofblocks,
+			XFS_ICI_COWBLOCKS_TAG);
+}
+
+void
+xfs_inode_clear_cowblocks_tag(
+	xfs_inode_t	*ip)
+{
+	trace_xfs_inode_clear_eofblocks_tag(ip);
+	return __xfs_inode_clear_eofblocks_tag(ip,
+			trace_xfs_perag_clear_eofblocks, XFS_ICI_COWBLOCKS_TAG);
+}

fs/xfs/xfs_icache.h

@@ -40,6 +40,7 @@ struct xfs_eofblocks {
 					   in xfs_inode_ag_iterator */
 #define XFS_ICI_RECLAIM_TAG	0	/* inode is to be reclaimed */
 #define XFS_ICI_EOFBLOCKS_TAG	1	/* inode has blocks beyond EOF */
+#define XFS_ICI_COWBLOCKS_TAG	2	/* inode can have cow blocks to gc */
 
 /*
  * Flags for xfs_iget()
@@ -70,6 +71,12 @@ int xfs_inode_free_quota_eofblocks(struct xfs_inode *ip);
 void xfs_eofblocks_worker(struct work_struct *);
 void xfs_queue_eofblocks(struct xfs_mount *);
 
+void xfs_inode_set_cowblocks_tag(struct xfs_inode *ip);
+void xfs_inode_clear_cowblocks_tag(struct xfs_inode *ip);
+int xfs_icache_free_cowblocks(struct xfs_mount *, struct xfs_eofblocks *);
+int xfs_inode_free_quota_cowblocks(struct xfs_inode *ip);
+void xfs_cowblocks_worker(struct work_struct *);
+
 int xfs_inode_ag_iterator(struct xfs_mount *mp,
 	int (*execute)(struct xfs_inode *ip, int flags, void *args),
 	int flags, void *args);

fs/xfs/xfs_inode.c

@@ -49,6 +49,7 @@
 #include "xfs_trans_priv.h"
 #include "xfs_log.h"
 #include "xfs_bmap_btree.h"
+#include "xfs_reflink.h"
 
 kmem_zone_t *xfs_inode_zone;
 
@@ -76,6 +77,29 @@ xfs_get_extsz_hint(
 	return 0;
 }
 
+/*
+ * Helper function to extract CoW extent size hint from inode.
+ * Between the extent size hint and the CoW extent size hint, we
+ * return the greater of the two.  If the value is zero (automatic),
+ * use the default size.
+ */
+xfs_extlen_t
+xfs_get_cowextsz_hint(
+	struct xfs_inode	*ip)
+{
+	xfs_extlen_t		a, b;
+
+	a = 0;
+	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
+		a = ip->i_d.di_cowextsize;
+	b = xfs_get_extsz_hint(ip);
+
+	a = max(a, b);
+	if (a == 0)
+		return XFS_DEFAULT_COWEXTSZ_HINT;
+	return a;
+}
+
 /*
  * These two are wrapper routines around the xfs_ilock() routine used to
  * centralize some grungy code.  They are used in places that wish to lock the
@@ -651,6 +675,8 @@ _xfs_dic2xflags(
 	if (di_flags2 & XFS_DIFLAG2_ANY) {
 		if (di_flags2 & XFS_DIFLAG2_DAX)
 			flags |= FS_XFLAG_DAX;
+		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
+			flags |= FS_XFLAG_COWEXTSIZE;
 	}
 
 	if (has_attr)
@@ -834,6 +860,7 @@ xfs_ialloc(
 	if (ip->i_d.di_version == 3) {
 		inode->i_version = 1;
 		ip->i_d.di_flags2 = 0;
+		ip->i_d.di_cowextsize = 0;
 		ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
 		ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
 	}
@@ -896,6 +923,15 @@ xfs_ialloc(
 			ip->i_d.di_flags |= di_flags;
 			ip->i_d.di_flags2 |= di_flags2;
 		}
+		if (pip &&
+		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
+		    pip->i_d.di_version == 3 &&
+		    ip->i_d.di_version == 3) {
+			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
+				ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
+				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
+			}
+		}
 		/* FALLTHROUGH */
 	case S_IFLNK:
 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
@@ -1586,6 +1622,20 @@ xfs_itruncate_extents(
 			goto out;
 	}
 
+	/* Remove all pending CoW reservations. */
+	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
+			last_block);
+	if (error)
+		goto out;
+
+	/*
+	 * Clear the reflink flag if we truncated everything.
+	 */
+	if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) {
+		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+		xfs_inode_clear_cowblocks_tag(ip);
+	}
+
 	/*
 	 * Always re-log the inode so that our permanent transaction can keep
 	 * on rolling it forward in the log.
@@ -1850,6 +1900,7 @@ xfs_inactive(
 	}
 
 	mp = ip->i_mount;
+	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
 
 	/* If this is a read-only mount, don't do this (would generate I/O) */
 	if (mp->m_flags & XFS_MOUNT_RDONLY)

fs/xfs/xfs_inode.h

@@ -47,6 +47,7 @@ typedef struct xfs_inode {
 
 	/* Extent information. */
 	xfs_ifork_t		*i_afp;		/* attribute fork pointer */
+	xfs_ifork_t		*i_cowfp;	/* copy on write extents */
 	xfs_ifork_t		i_df;		/* data fork */
 
 	/* operations vectors */
@@ -65,6 +66,9 @@ typedef struct xfs_inode {
 
 	struct xfs_icdinode	i_d;		/* most of ondisk inode */
 
+	xfs_extnum_t		i_cnextents;	/* # of extents in cow fork */
+	unsigned int		i_cformat;	/* format of cow fork */
+
 	/* VFS inode */
 	struct inode		i_vnode;	/* embedded VFS inode */
 } xfs_inode_t;
@@ -202,6 +206,11 @@ xfs_get_initial_prid(struct xfs_inode *dp)
 	return XFS_PROJID_DEFAULT;
 }
 
+static inline bool xfs_is_reflink_inode(struct xfs_inode *ip)
+{
+	return ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
+}
+
 /*
  * In-core inode flags.
  */
@@ -217,6 +226,12 @@ xfs_get_initial_prid(struct xfs_inode *dp)
 #define XFS_IPINNED		(1 << __XFS_IPINNED_BIT)
 #define XFS_IDONTCACHE		(1 << 9) /* don't cache the inode long term */
 #define XFS_IEOFBLOCKS		(1 << 10)/* has the preallocblocks tag set */
+/*
+ * If this unlinked inode is in the middle of recovery, don't let drop_inode
+ * truncate and free the inode.  This can happen if we iget the inode during
+ * log recovery to replay a bmap operation on the inode.
+ */
+#define XFS_IRECOVERY		(1 << 11)
 
 /*
  * Per-lifetime flags need to be reset when re-using a reclaimable inode during
@@ -411,6 +426,7 @@ int		xfs_iflush(struct xfs_inode *, struct xfs_buf **);
 void		xfs_lock_two_inodes(xfs_inode_t *, xfs_inode_t *, uint);
 
 xfs_extlen_t	xfs_get_extsz_hint(struct xfs_inode *ip);
+xfs_extlen_t	xfs_get_cowextsz_hint(struct xfs_inode *ip);
 
 int		xfs_dir_ialloc(struct xfs_trans **, struct xfs_inode *, umode_t,
 			       xfs_nlink_t, xfs_dev_t, prid_t, int,
@@ -474,4 +490,7 @@ do { \
 
 extern struct kmem_zone	*xfs_inode_zone;
 
+/* The default CoW extent size hint. */
+#define XFS_DEFAULT_COWEXTSZ_HINT 32
+
 #endif	/* __XFS_INODE_H__ */

fs/xfs/xfs_inode_item.c

@@ -368,7 +368,7 @@ xfs_inode_to_log_dinode(
 		to->di_crtime.t_sec = from->di_crtime.t_sec;
 		to->di_crtime.t_nsec = from->di_crtime.t_nsec;
 		to->di_flags2 = from->di_flags2;
-
+		to->di_cowextsize = from->di_cowextsize;
 		to->di_ino = ip->i_ino;
 		to->di_lsn = lsn;
 		memset(to->di_pad2, 0, sizeof(to->di_pad2));

fs/xfs/xfs_ioctl.c

@@ -903,6 +903,8 @@ xfs_ioc_fsgetxattr(
 	xfs_ilock(ip, XFS_ILOCK_SHARED);
 	fa.fsx_xflags = xfs_ip2xflags(ip);
 	fa.fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
+	fa.fsx_cowextsize = ip->i_d.di_cowextsize <<
+			ip->i_mount->m_sb.sb_blocklog;
 	fa.fsx_projid = xfs_get_projid(ip);
 
 	if (attr) {
@@ -973,12 +975,13 @@ xfs_set_diflags(
 	if (ip->i_d.di_version < 3)
 		return;
 
-	di_flags2 = 0;
+	di_flags2 = (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK);
 	if (xflags & FS_XFLAG_DAX)
 		di_flags2 |= XFS_DIFLAG2_DAX;
+	if (xflags & FS_XFLAG_COWEXTSIZE)
+		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
 
 	ip->i_d.di_flags2 = di_flags2;
-
 }
 
 STATIC void
@@ -1031,6 +1034,14 @@ xfs_ioctl_setattr_xflags(
 			return -EINVAL;
 	}
 
+	/* Clear reflink if we are actually able to set the rt flag. */
+	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
+		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+
+	/* Don't allow us to set DAX mode for a reflinked file for now. */
+	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
+		return -EINVAL;
+
 	/*
 	 * Can't modify an immutable/append-only file unless
 	 * we have appropriate permission.
@@ -1219,6 +1230,56 @@ xfs_ioctl_setattr_check_extsize(
 	return 0;
 }
 
+/*
+ * CoW extent size hint validation rules are:
+ *
+ * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
+ *    The inode does not have to have any shared blocks, but it must be a v3.
+ * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
+ *    for a directory, the hint is propagated to new files.
+ * 3. Can be changed on files & directories at any time.
+ * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
+ * 5. Extent size must be a multiple of the appropriate block size.
+ * 6. The extent size hint must be limited to half the AG size to avoid
+ *    alignment extending the extent beyond the limits of the AG.
+ */
+static int
+xfs_ioctl_setattr_check_cowextsize(
+	struct xfs_inode	*ip,
+	struct fsxattr		*fa)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+
+	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
+		return 0;
+
+	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb) ||
+	    ip->i_d.di_version != 3)
+		return -EINVAL;
+
+	if (!S_ISREG(VFS_I(ip)->i_mode) && !S_ISDIR(VFS_I(ip)->i_mode))
+		return -EINVAL;
+
+	if (fa->fsx_cowextsize != 0) {
+		xfs_extlen_t    size;
+		xfs_fsblock_t   cowextsize_fsb;
+
+		cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
+		if (cowextsize_fsb > MAXEXTLEN)
+			return -EINVAL;
+
+		size = mp->m_sb.sb_blocksize;
+		if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
+			return -EINVAL;
+
+		if (fa->fsx_cowextsize % size)
+			return -EINVAL;
+	} else
+		fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
+
+	return 0;
+}
+
 static int
 xfs_ioctl_setattr_check_projid(
 	struct xfs_inode	*ip,
@@ -1311,6 +1372,10 @@ xfs_ioctl_setattr(
 	if (code)
 		goto error_trans_cancel;
 
+	code = xfs_ioctl_setattr_check_cowextsize(ip, fa);
+	if (code)
+		goto error_trans_cancel;
+
 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
 	if (code)
 		goto error_trans_cancel;
@@ -1346,6 +1411,12 @@ xfs_ioctl_setattr(
 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
 	else
 		ip->i_d.di_extsize = 0;
+	if (ip->i_d.di_version == 3 &&
+	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
+		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
+				mp->m_sb.sb_blocklog;
+	else
+		ip->i_d.di_cowextsize = 0;
 
 	code = xfs_trans_commit(tp);
 

fs/xfs/xfs_iomap.c

@@ -39,6 +39,7 @@
 #include "xfs_quota.h"
 #include "xfs_dquot_item.h"
 #include "xfs_dquot.h"
+#include "xfs_reflink.h"
 
 
 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
@@ -70,7 +71,7 @@ xfs_bmbt_to_iomap(
 	iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
 }
 
-static xfs_extlen_t
+xfs_extlen_t
 xfs_eof_alignment(
 	struct xfs_inode	*ip,
 	xfs_extlen_t		extsize)
@@ -609,7 +610,7 @@ xfs_file_iomap_begin_delay(
 	}
 
 retry:
-	error = xfs_bmapi_reserve_delalloc(ip, offset_fsb,
+	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
 			end_fsb - offset_fsb, &got,
 			&prev, &idx, eof);
 	switch (error) {
@@ -666,6 +667,7 @@ out_unlock:
 int
 xfs_iomap_write_allocate(
 	xfs_inode_t	*ip,
+	int		whichfork,
 	xfs_off_t	offset,
 	xfs_bmbt_irec_t *imap)
 {
@@ -678,8 +680,12 @@ xfs_iomap_write_allocate(
 	xfs_trans_t	*tp;
 	int		nimaps;
 	int		error = 0;
+	int		flags = 0;
 	int		nres;
 
+	if (whichfork == XFS_COW_FORK)
+		flags |= XFS_BMAPI_COWFORK;
+
 	/*
 	 * Make sure that the dquots are there.
 	 */
@@ -773,7 +779,7 @@ xfs_iomap_write_allocate(
 			 * pointer that the caller gave to us.
 			 */
 			error = xfs_bmapi_write(tp, ip, map_start_fsb,
-						count_fsb, 0, &first_block,
+						count_fsb, flags, &first_block,
 						nres, imap, &nimaps,
 						&dfops);
 			if (error)
@@ -955,14 +961,22 @@ xfs_file_iomap_begin(
 	struct xfs_mount	*mp = ip->i_mount;
 	struct xfs_bmbt_irec	imap;
 	xfs_fileoff_t		offset_fsb, end_fsb;
+	bool			shared, trimmed;
 	int			nimaps = 1, error = 0;
 	unsigned		lockmode;
 
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return -EIO;
 
-	if ((flags & IOMAP_WRITE) &&
-	    !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
+	if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && xfs_is_reflink_inode(ip)) {
+		error = xfs_reflink_reserve_cow_range(ip, offset, length);
+		if (error < 0)
+			return error;
+	}
+
+	if ((flags & IOMAP_WRITE) && !IS_DAX(inode) &&
+		   !xfs_get_extsz_hint(ip)) {
+		/* Reserve delalloc blocks for regular writeback. */
 		return xfs_file_iomap_begin_delay(inode, offset, length, flags,
 				iomap);
 	}
@@ -976,7 +990,14 @@ xfs_file_iomap_begin(
 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
 
 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
-			       &nimaps, XFS_BMAPI_ENTIRE);
+			       &nimaps, 0);
+	if (error) {
+		xfs_iunlock(ip, lockmode);
+		return error;
+	}
+
+	/* Trim the mapping to the nearest shared extent boundary. */
+	error = xfs_reflink_trim_around_shared(ip, &imap, &shared, &trimmed);
 	if (error) {
 		xfs_iunlock(ip, lockmode);
 		return error;
@@ -1015,6 +1036,8 @@ xfs_file_iomap_begin(
 	}
 
 	xfs_bmbt_to_iomap(ip, iomap, &imap);
+	if (shared)
+		iomap->flags |= IOMAP_F_SHARED;
 	return 0;
 }
 

fs/xfs/xfs_iomap.h

@@ -25,12 +25,13 @@ struct xfs_bmbt_irec;
 
 int xfs_iomap_write_direct(struct xfs_inode *, xfs_off_t, size_t,
 			struct xfs_bmbt_irec *, int);
-int xfs_iomap_write_allocate(struct xfs_inode *, xfs_off_t,
+int xfs_iomap_write_allocate(struct xfs_inode *, int, xfs_off_t,
 			struct xfs_bmbt_irec *);
 int xfs_iomap_write_unwritten(struct xfs_inode *, xfs_off_t, xfs_off_t);
 
 void xfs_bmbt_to_iomap(struct xfs_inode *, struct iomap *,
 		struct xfs_bmbt_irec *);
+xfs_extlen_t xfs_eof_alignment(struct xfs_inode *ip, xfs_extlen_t extsize);
 
 extern struct iomap_ops xfs_iomap_ops;
 extern struct iomap_ops xfs_xattr_iomap_ops;

fs/xfs/xfs_iops.c

@@ -1159,6 +1159,7 @@ xfs_diflags_to_iflags(
 		inode->i_flags |= S_NOATIME;
 	if (S_ISREG(inode->i_mode) &&
 	    ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
+	    !xfs_is_reflink_inode(ip) &&
 	    (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
 	     ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
 		inode->i_flags |= S_DAX;

fs/xfs/xfs_itable.c

@@ -66,7 +66,7 @@ xfs_bulkstat_one_int(
 	if (!buffer || xfs_internal_inum(mp, ino))
 		return -EINVAL;
 
-	buf = kmem_alloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
+	buf = kmem_zalloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
 	if (!buf)
 		return -ENOMEM;
 
@@ -111,6 +111,12 @@ xfs_bulkstat_one_int(
 	buf->bs_aextents = dic->di_anextents;
 	buf->bs_forkoff = XFS_IFORK_BOFF(ip);
 
+	if (dic->di_version == 3) {
+		if (dic->di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
+			buf->bs_cowextsize = dic->di_cowextsize <<
+					mp->m_sb.sb_blocklog;
+	}
+
 	switch (dic->di_format) {
 	case XFS_DINODE_FMT_DEV:
 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;

fs/xfs/xfs_linux.h

@@ -116,6 +116,7 @@ typedef __u32			xfs_nlink_t;
 #define xfs_inherit_nodefrag	xfs_params.inherit_nodfrg.val
 #define xfs_fstrm_centisecs	xfs_params.fstrm_timer.val
 #define xfs_eofb_secs		xfs_params.eofb_timer.val
+#define xfs_cowb_secs		xfs_params.cowb_timer.val
 
 #define current_cpu()		(raw_smp_processor_id())
 #define current_pid()		(current->pid)

fs/xfs/xfs_log_recover.c

@@ -45,6 +45,8 @@
 #include "xfs_dir2.h"
 #include "xfs_rmap_item.h"
 #include "xfs_buf_item.h"
+#include "xfs_refcount_item.h"
+#include "xfs_bmap_item.h"
 
 #define BLK_AVG(blk1, blk2)	((blk1+blk2) >> 1)
 
@@ -1924,6 +1926,10 @@ xlog_recover_reorder_trans(
 		case XFS_LI_EFI:
 		case XFS_LI_RUI:
 		case XFS_LI_RUD:
+		case XFS_LI_CUI:
+		case XFS_LI_CUD:
+		case XFS_LI_BUI:
+		case XFS_LI_BUD:
 			trace_xfs_log_recover_item_reorder_tail(log,
 							trans, item, pass);
 			list_move_tail(&item->ri_list, &inode_list);
@@ -2242,6 +2248,7 @@ xlog_recover_get_buf_lsn(
 	case XFS_ABTB_MAGIC:
 	case XFS_ABTC_MAGIC:
 	case XFS_RMAP_CRC_MAGIC:
+	case XFS_REFC_CRC_MAGIC:
 	case XFS_IBT_CRC_MAGIC:
 	case XFS_IBT_MAGIC: {
 		struct xfs_btree_block *btb = blk;
@@ -2415,6 +2422,9 @@ xlog_recover_validate_buf_type(
 		case XFS_RMAP_CRC_MAGIC:
 			bp->b_ops = &xfs_rmapbt_buf_ops;
 			break;
+		case XFS_REFC_CRC_MAGIC:
+			bp->b_ops = &xfs_refcountbt_buf_ops;
+			break;
 		default:
 			warnmsg = "Bad btree block magic!";
 			break;
@@ -3546,6 +3556,242 @@ xlog_recover_rud_pass2(
 	return 0;
 }
 
+/*
+ * Copy an CUI format buffer from the given buf, and into the destination
+ * CUI format structure.  The CUI/CUD items were designed not to need any
+ * special alignment handling.
+ */
+static int
+xfs_cui_copy_format(
+	struct xfs_log_iovec		*buf,
+	struct xfs_cui_log_format	*dst_cui_fmt)
+{
+	struct xfs_cui_log_format	*src_cui_fmt;
+	uint				len;
+
+	src_cui_fmt = buf->i_addr;
+	len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
+
+	if (buf->i_len == len) {
+		memcpy(dst_cui_fmt, src_cui_fmt, len);
+		return 0;
+	}
+	return -EFSCORRUPTED;
+}
+
+/*
+ * This routine is called to create an in-core extent refcount update
+ * item from the cui format structure which was logged on disk.
+ * It allocates an in-core cui, copies the extents from the format
+ * structure into it, and adds the cui to the AIL with the given
+ * LSN.
+ */
+STATIC int
+xlog_recover_cui_pass2(
+	struct xlog			*log,
+	struct xlog_recover_item	*item,
+	xfs_lsn_t			lsn)
+{
+	int				error;
+	struct xfs_mount		*mp = log->l_mp;
+	struct xfs_cui_log_item		*cuip;
+	struct xfs_cui_log_format	*cui_formatp;
+
+	cui_formatp = item->ri_buf[0].i_addr;
+
+	cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
+	error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
+	if (error) {
+		xfs_cui_item_free(cuip);
+		return error;
+	}
+	atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
+
+	spin_lock(&log->l_ailp->xa_lock);
+	/*
+	 * The CUI has two references. One for the CUD and one for CUI to ensure
+	 * it makes it into the AIL. Insert the CUI into the AIL directly and
+	 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
+	 * AIL lock.
+	 */
+	xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
+	xfs_cui_release(cuip);
+	return 0;
+}
+
+
+/*
+ * This routine is called when an CUD format structure is found in a committed
+ * transaction in the log. Its purpose is to cancel the corresponding CUI if it
+ * was still in the log. To do this it searches the AIL for the CUI with an id
+ * equal to that in the CUD format structure. If we find it we drop the CUD
+ * reference, which removes the CUI from the AIL and frees it.
+ */
+STATIC int
+xlog_recover_cud_pass2(
+	struct xlog			*log,
+	struct xlog_recover_item	*item)
+{
+	struct xfs_cud_log_format	*cud_formatp;
+	struct xfs_cui_log_item		*cuip = NULL;
+	struct xfs_log_item		*lip;
+	__uint64_t			cui_id;
+	struct xfs_ail_cursor		cur;
+	struct xfs_ail			*ailp = log->l_ailp;
+
+	cud_formatp = item->ri_buf[0].i_addr;
+	if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format))
+		return -EFSCORRUPTED;
+	cui_id = cud_formatp->cud_cui_id;
+
+	/*
+	 * Search for the CUI with the id in the CUD format structure in the
+	 * AIL.
+	 */
+	spin_lock(&ailp->xa_lock);
+	lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
+	while (lip != NULL) {
+		if (lip->li_type == XFS_LI_CUI) {
+			cuip = (struct xfs_cui_log_item *)lip;
+			if (cuip->cui_format.cui_id == cui_id) {
+				/*
+				 * Drop the CUD reference to the CUI. This
+				 * removes the CUI from the AIL and frees it.
+				 */
+				spin_unlock(&ailp->xa_lock);
+				xfs_cui_release(cuip);
+				spin_lock(&ailp->xa_lock);
+				break;
+			}
+		}
+		lip = xfs_trans_ail_cursor_next(ailp, &cur);
+	}
+
+	xfs_trans_ail_cursor_done(&cur);
+	spin_unlock(&ailp->xa_lock);
+
+	return 0;
+}
+
+/*
+ * Copy an BUI format buffer from the given buf, and into the destination
+ * BUI format structure.  The BUI/BUD items were designed not to need any
+ * special alignment handling.
+ */
+static int
+xfs_bui_copy_format(
+	struct xfs_log_iovec		*buf,
+	struct xfs_bui_log_format	*dst_bui_fmt)
+{
+	struct xfs_bui_log_format	*src_bui_fmt;
+	uint				len;
+
+	src_bui_fmt = buf->i_addr;
+	len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
+
+	if (buf->i_len == len) {
+		memcpy(dst_bui_fmt, src_bui_fmt, len);
+		return 0;
+	}
+	return -EFSCORRUPTED;
+}
+
+/*
+ * This routine is called to create an in-core extent bmap update
+ * item from the bui format structure which was logged on disk.
+ * It allocates an in-core bui, copies the extents from the format
+ * structure into it, and adds the bui to the AIL with the given
+ * LSN.
+ */
+STATIC int
+xlog_recover_bui_pass2(
+	struct xlog			*log,
+	struct xlog_recover_item	*item,
+	xfs_lsn_t			lsn)
+{
+	int				error;
+	struct xfs_mount		*mp = log->l_mp;
+	struct xfs_bui_log_item		*buip;
+	struct xfs_bui_log_format	*bui_formatp;
+
+	bui_formatp = item->ri_buf[0].i_addr;
+
+	if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
+		return -EFSCORRUPTED;
+	buip = xfs_bui_init(mp);
+	error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
+	if (error) {
+		xfs_bui_item_free(buip);
+		return error;
+	}
+	atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
+
+	spin_lock(&log->l_ailp->xa_lock);
+	/*
+	 * The RUI has two references. One for the RUD and one for RUI to ensure
+	 * it makes it into the AIL. Insert the RUI into the AIL directly and
+	 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
+	 * AIL lock.
+	 */
+	xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
+	xfs_bui_release(buip);
+	return 0;
+}
+
+
+/*
+ * This routine is called when an BUD format structure is found in a committed
+ * transaction in the log. Its purpose is to cancel the corresponding BUI if it
+ * was still in the log. To do this it searches the AIL for the BUI with an id
+ * equal to that in the BUD format structure. If we find it we drop the BUD
+ * reference, which removes the BUI from the AIL and frees it.
+ */
+STATIC int
+xlog_recover_bud_pass2(
+	struct xlog			*log,
+	struct xlog_recover_item	*item)
+{
+	struct xfs_bud_log_format	*bud_formatp;
+	struct xfs_bui_log_item		*buip = NULL;
+	struct xfs_log_item		*lip;
+	__uint64_t			bui_id;
+	struct xfs_ail_cursor		cur;
+	struct xfs_ail			*ailp = log->l_ailp;
+
+	bud_formatp = item->ri_buf[0].i_addr;
+	if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format))
+		return -EFSCORRUPTED;
+	bui_id = bud_formatp->bud_bui_id;
+
+	/*
+	 * Search for the BUI with the id in the BUD format structure in the
+	 * AIL.
+	 */
+	spin_lock(&ailp->xa_lock);
+	lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
+	while (lip != NULL) {
+		if (lip->li_type == XFS_LI_BUI) {
+			buip = (struct xfs_bui_log_item *)lip;
+			if (buip->bui_format.bui_id == bui_id) {
+				/*
+				 * Drop the BUD reference to the BUI. This
+				 * removes the BUI from the AIL and frees it.
+				 */
+				spin_unlock(&ailp->xa_lock);
+				xfs_bui_release(buip);
+				spin_lock(&ailp->xa_lock);
+				break;
+			}
+		}
+		lip = xfs_trans_ail_cursor_next(ailp, &cur);
+	}
+
+	xfs_trans_ail_cursor_done(&cur);
+	spin_unlock(&ailp->xa_lock);
+
+	return 0;
+}
+
 /*
  * This routine is called when an inode create format structure is found in a
  * committed transaction in the log.  It's purpose is to initialise the inodes
@@ -3773,6 +4019,10 @@ xlog_recover_ra_pass2(
 	case XFS_LI_QUOTAOFF:
 	case XFS_LI_RUI:
 	case XFS_LI_RUD:
+	case XFS_LI_CUI:
+	case XFS_LI_CUD:
+	case XFS_LI_BUI:
+	case XFS_LI_BUD:
 	default:
 		break;
 	}
@@ -3798,6 +4048,10 @@ xlog_recover_commit_pass1(
 	case XFS_LI_ICREATE:
 	case XFS_LI_RUI:
 	case XFS_LI_RUD:
+	case XFS_LI_CUI:
+	case XFS_LI_CUD:
+	case XFS_LI_BUI:
+	case XFS_LI_BUD:
 		/* nothing to do in pass 1 */
 		return 0;
 	default:
@@ -3832,6 +4086,14 @@ xlog_recover_commit_pass2(
 		return xlog_recover_rui_pass2(log, item, trans->r_lsn);
 	case XFS_LI_RUD:
 		return xlog_recover_rud_pass2(log, item);
+	case XFS_LI_CUI:
+		return xlog_recover_cui_pass2(log, item, trans->r_lsn);
+	case XFS_LI_CUD:
+		return xlog_recover_cud_pass2(log, item);
+	case XFS_LI_BUI:
+		return xlog_recover_bui_pass2(log, item, trans->r_lsn);
+	case XFS_LI_BUD:
+		return xlog_recover_bud_pass2(log, item);
 	case XFS_LI_DQUOT:
 		return xlog_recover_dquot_pass2(log, buffer_list, item,
 						trans->r_lsn);
@@ -4419,12 +4681,94 @@ xlog_recover_cancel_rui(
 	spin_lock(&ailp->xa_lock);
 }
 
+/* Recover the CUI if necessary. */
+STATIC int
+xlog_recover_process_cui(
+	struct xfs_mount		*mp,
+	struct xfs_ail			*ailp,
+	struct xfs_log_item		*lip)
+{
+	struct xfs_cui_log_item		*cuip;
+	int				error;
+
+	/*
+	 * Skip CUIs that we've already processed.
+	 */
+	cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
+	if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
+		return 0;
+
+	spin_unlock(&ailp->xa_lock);
+	error = xfs_cui_recover(mp, cuip);
+	spin_lock(&ailp->xa_lock);
+
+	return error;
+}
+
+/* Release the CUI since we're cancelling everything. */
+STATIC void
+xlog_recover_cancel_cui(
+	struct xfs_mount		*mp,
+	struct xfs_ail			*ailp,
+	struct xfs_log_item		*lip)
+{
+	struct xfs_cui_log_item		*cuip;
+
+	cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
+
+	spin_unlock(&ailp->xa_lock);
+	xfs_cui_release(cuip);
+	spin_lock(&ailp->xa_lock);
+}
+
+/* Recover the BUI if necessary. */
+STATIC int
+xlog_recover_process_bui(
+	struct xfs_mount		*mp,
+	struct xfs_ail			*ailp,
+	struct xfs_log_item		*lip)
+{
+	struct xfs_bui_log_item		*buip;
+	int				error;
+
+	/*
+	 * Skip BUIs that we've already processed.
+	 */
+	buip = container_of(lip, struct xfs_bui_log_item, bui_item);
+	if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
+		return 0;
+
+	spin_unlock(&ailp->xa_lock);
+	error = xfs_bui_recover(mp, buip);
+	spin_lock(&ailp->xa_lock);
+
+	return error;
+}
+
+/* Release the BUI since we're cancelling everything. */
+STATIC void
+xlog_recover_cancel_bui(
+	struct xfs_mount		*mp,
+	struct xfs_ail			*ailp,
+	struct xfs_log_item		*lip)
+{
+	struct xfs_bui_log_item		*buip;
+
+	buip = container_of(lip, struct xfs_bui_log_item, bui_item);
+
+	spin_unlock(&ailp->xa_lock);
+	xfs_bui_release(buip);
+	spin_lock(&ailp->xa_lock);
+}
+
 /* Is this log item a deferred action intent? */
 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
 {
 	switch (lip->li_type) {
 	case XFS_LI_EFI:
 	case XFS_LI_RUI:
+	case XFS_LI_CUI:
+	case XFS_LI_BUI:
 		return true;
 	default:
 		return false;
@@ -4488,6 +4832,12 @@ xlog_recover_process_intents(
 		case XFS_LI_RUI:
 			error = xlog_recover_process_rui(log->l_mp, ailp, lip);
 			break;
+		case XFS_LI_CUI:
+			error = xlog_recover_process_cui(log->l_mp, ailp, lip);
+			break;
+		case XFS_LI_BUI:
+			error = xlog_recover_process_bui(log->l_mp, ailp, lip);
+			break;
 		}
 		if (error)
 			goto out;
@@ -4535,6 +4885,12 @@ xlog_recover_cancel_intents(
 		case XFS_LI_RUI:
 			xlog_recover_cancel_rui(log->l_mp, ailp, lip);
 			break;
+		case XFS_LI_CUI:
+			xlog_recover_cancel_cui(log->l_mp, ailp, lip);
+			break;
+		case XFS_LI_BUI:
+			xlog_recover_cancel_bui(log->l_mp, ailp, lip);
+			break;
 		}
 
 		lip = xfs_trans_ail_cursor_next(ailp, &cur);
@@ -4613,6 +4969,7 @@ xlog_recover_process_one_iunlink(
 	if (error)
 		goto fail_iput;
 
+	xfs_iflags_clear(ip, XFS_IRECOVERY);
 	ASSERT(VFS_I(ip)->i_nlink == 0);
 	ASSERT(VFS_I(ip)->i_mode != 0);
 

fs/xfs/xfs_mount.c

@@ -43,6 +43,8 @@
 #include "xfs_icache.h"
 #include "xfs_sysfs.h"
 #include "xfs_rmap_btree.h"
+#include "xfs_refcount_btree.h"
+#include "xfs_reflink.h"
 
 
 static DEFINE_MUTEX(xfs_uuid_table_mutex);
@@ -684,6 +686,7 @@ xfs_mountfs(
 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 	xfs_ialloc_compute_maxlevels(mp);
 	xfs_rmapbt_compute_maxlevels(mp);
+	xfs_refcountbt_compute_maxlevels(mp);
 
 	xfs_set_maxicount(mp);
 
@@ -922,6 +925,15 @@ xfs_mountfs(
 		}
 	}
 
+	/*
+	 * During the second phase of log recovery, we need iget and
+	 * iput to behave like they do for an active filesystem.
+	 * xfs_fs_drop_inode needs to be able to prevent the deletion
+	 * of inodes before we're done replaying log items on those
+	 * inodes.
+	 */
+	mp->m_super->s_flags |= MS_ACTIVE;
+
 	/*
 	 * Finish recovering the file system.  This part needed to be delayed
 	 * until after the root and real-time bitmap inodes were consistently
@@ -974,10 +986,28 @@ xfs_mountfs(
 		if (error)
 			xfs_warn(mp,
 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
+
+		/* Recover any CoW blocks that never got remapped. */
+		error = xfs_reflink_recover_cow(mp);
+		if (error) {
+			xfs_err(mp,
+	"Error %d recovering leftover CoW allocations.", error);
+			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+			goto out_quota;
+		}
+
+		/* Reserve AG blocks for future btree expansion. */
+		error = xfs_fs_reserve_ag_blocks(mp);
+		if (error && error != -ENOSPC)
+			goto out_agresv;
 	}
 
 	return 0;
 
+ out_agresv:
+	xfs_fs_unreserve_ag_blocks(mp);
+ out_quota:
+	xfs_qm_unmount_quotas(mp);
  out_rtunmount:
 	xfs_rtunmount_inodes(mp);
  out_rele_rip:
@@ -1019,7 +1049,9 @@ xfs_unmountfs(
 	int			error;
 
 	cancel_delayed_work_sync(&mp->m_eofblocks_work);
+	cancel_delayed_work_sync(&mp->m_cowblocks_work);
 
+	xfs_fs_unreserve_ag_blocks(mp);
 	xfs_qm_unmount_quotas(mp);
 	xfs_rtunmount_inodes(mp);
 	IRELE(mp->m_rootip);

fs/xfs/xfs_mount.h

@@ -124,10 +124,13 @@ typedef struct xfs_mount {
 	uint			m_inobt_mnr[2];	/* min inobt btree records */
 	uint			m_rmap_mxr[2];	/* max rmap btree records */
 	uint			m_rmap_mnr[2];	/* min rmap btree records */
+	uint			m_refc_mxr[2];	/* max refc btree records */
+	uint			m_refc_mnr[2];	/* min refc btree records */
 	uint			m_ag_maxlevels;	/* XFS_AG_MAXLEVELS */
 	uint			m_bm_maxlevels[2]; /* XFS_BM_MAXLEVELS */
 	uint			m_in_maxlevels;	/* max inobt btree levels. */
 	uint			m_rmap_maxlevels; /* max rmap btree levels */
+	uint			m_refc_maxlevels; /* max refcount btree level */
 	xfs_extlen_t		m_ag_prealloc_blocks; /* reserved ag blocks */
 	uint			m_alloc_set_aside; /* space we can't use */
 	uint			m_ag_max_usable; /* max space per AG */
@@ -161,6 +164,8 @@ typedef struct xfs_mount {
 	struct delayed_work	m_reclaim_work;	/* background inode reclaim */
 	struct delayed_work	m_eofblocks_work; /* background eof blocks
 						     trimming */
+	struct delayed_work	m_cowblocks_work; /* background cow blocks
+						     trimming */
 	bool			m_update_sb;	/* sb needs update in mount */
 	int64_t			m_low_space[XFS_LOWSP_MAX];
 						/* low free space thresholds */
@@ -399,6 +404,9 @@ typedef struct xfs_perag {
 	struct xfs_ag_resv	pag_meta_resv;
 	/* Blocks reserved for just AGFL-based metadata. */
 	struct xfs_ag_resv	pag_agfl_resv;
+
+	/* reference count */
+	__uint8_t		pagf_refcount_level;
 } xfs_perag_t;
 
 static inline struct xfs_ag_resv *

fs/xfs/xfs_ondisk.h

@@ -49,6 +49,8 @@ xfs_check_ondisk_structs(void)
 	XFS_CHECK_STRUCT_SIZE(struct xfs_dsymlink_hdr,		56);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_inobt_key,		4);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_inobt_rec,		16);
+	XFS_CHECK_STRUCT_SIZE(struct xfs_refcount_key,		4);
+	XFS_CHECK_STRUCT_SIZE(struct xfs_refcount_rec,		12);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_rmap_key,		20);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_rmap_rec,		24);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_timestamp,		8);
@@ -56,6 +58,7 @@ xfs_check_ondisk_structs(void)
 	XFS_CHECK_STRUCT_SIZE(xfs_alloc_ptr_t,			4);
 	XFS_CHECK_STRUCT_SIZE(xfs_alloc_rec_t,			8);
 	XFS_CHECK_STRUCT_SIZE(xfs_inobt_ptr_t,			4);
+	XFS_CHECK_STRUCT_SIZE(xfs_refcount_ptr_t,		4);
 	XFS_CHECK_STRUCT_SIZE(xfs_rmap_ptr_t,			4);
 
 	/* dir/attr trees */

fs/xfs/xfs_pnfs.c

@@ -113,6 +113,13 @@ xfs_fs_map_blocks(
 	if (XFS_IS_REALTIME_INODE(ip))
 		return -ENXIO;
 
+	/*
+	 * The pNFS block layout spec actually supports reflink like
+	 * functionality, but the Linux pNFS server doesn't implement it yet.
+	 */
+	if (xfs_is_reflink_inode(ip))
+		return -ENXIO;
+
 	/*
 	 * Lock out any other I/O before we flush and invalidate the pagecache,
 	 * and then hand out a layout to the remote system.  This is very

fs/xfs/xfs_refcount_item.c

@@ -0,0 +1,539 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_bit.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_trans.h"
+#include "xfs_trans_priv.h"
+#include "xfs_buf_item.h"
+#include "xfs_refcount_item.h"
+#include "xfs_log.h"
+#include "xfs_refcount.h"
+
+
+kmem_zone_t	*xfs_cui_zone;
+kmem_zone_t	*xfs_cud_zone;
+
+static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
+{
+	return container_of(lip, struct xfs_cui_log_item, cui_item);
+}
+
+void
+xfs_cui_item_free(
+	struct xfs_cui_log_item	*cuip)
+{
+	if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
+		kmem_free(cuip);
+	else
+		kmem_zone_free(xfs_cui_zone, cuip);
+}
+
+STATIC void
+xfs_cui_item_size(
+	struct xfs_log_item	*lip,
+	int			*nvecs,
+	int			*nbytes)
+{
+	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
+
+	*nvecs += 1;
+	*nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
+}
+
+/*
+ * This is called to fill in the vector of log iovecs for the
+ * given cui log item. We use only 1 iovec, and we point that
+ * at the cui_log_format structure embedded in the cui item.
+ * It is at this point that we assert that all of the extent
+ * slots in the cui item have been filled.
+ */
+STATIC void
+xfs_cui_item_format(
+	struct xfs_log_item	*lip,
+	struct xfs_log_vec	*lv)
+{
+	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
+
+	ASSERT(atomic_read(&cuip->cui_next_extent) ==
+			cuip->cui_format.cui_nextents);
+
+	cuip->cui_format.cui_type = XFS_LI_CUI;
+	cuip->cui_format.cui_size = 1;
+
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format,
+			xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents));
+}
+
+/*
+ * Pinning has no meaning for an cui item, so just return.
+ */
+STATIC void
+xfs_cui_item_pin(
+	struct xfs_log_item	*lip)
+{
+}
+
+/*
+ * The unpin operation is the last place an CUI is manipulated in the log. It is
+ * either inserted in the AIL or aborted in the event of a log I/O error. In
+ * either case, the CUI transaction has been successfully committed to make it
+ * this far. Therefore, we expect whoever committed the CUI to either construct
+ * and commit the CUD or drop the CUD's reference in the event of error. Simply
+ * drop the log's CUI reference now that the log is done with it.
+ */
+STATIC void
+xfs_cui_item_unpin(
+	struct xfs_log_item	*lip,
+	int			remove)
+{
+	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
+
+	xfs_cui_release(cuip);
+}
+
+/*
+ * CUI items have no locking or pushing.  However, since CUIs are pulled from
+ * the AIL when their corresponding CUDs are committed to disk, their situation
+ * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
+ * will eventually flush the log.  This should help in getting the CUI out of
+ * the AIL.
+ */
+STATIC uint
+xfs_cui_item_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
+{
+	return XFS_ITEM_PINNED;
+}
+
+/*
+ * The CUI has been either committed or aborted if the transaction has been
+ * cancelled. If the transaction was cancelled, an CUD isn't going to be
+ * constructed and thus we free the CUI here directly.
+ */
+STATIC void
+xfs_cui_item_unlock(
+	struct xfs_log_item	*lip)
+{
+	if (lip->li_flags & XFS_LI_ABORTED)
+		xfs_cui_item_free(CUI_ITEM(lip));
+}
+
+/*
+ * The CUI is logged only once and cannot be moved in the log, so simply return
+ * the lsn at which it's been logged.
+ */
+STATIC xfs_lsn_t
+xfs_cui_item_committed(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+	return lsn;
+}
+
+/*
+ * The CUI dependency tracking op doesn't do squat.  It can't because
+ * it doesn't know where the free extent is coming from.  The dependency
+ * tracking has to be handled by the "enclosing" metadata object.  For
+ * example, for inodes, the inode is locked throughout the extent freeing
+ * so the dependency should be recorded there.
+ */
+STATIC void
+xfs_cui_item_committing(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+}
+
+/*
+ * This is the ops vector shared by all cui log items.
+ */
+static const struct xfs_item_ops xfs_cui_item_ops = {
+	.iop_size	= xfs_cui_item_size,
+	.iop_format	= xfs_cui_item_format,
+	.iop_pin	= xfs_cui_item_pin,
+	.iop_unpin	= xfs_cui_item_unpin,
+	.iop_unlock	= xfs_cui_item_unlock,
+	.iop_committed	= xfs_cui_item_committed,
+	.iop_push	= xfs_cui_item_push,
+	.iop_committing = xfs_cui_item_committing,
+};
+
+/*
+ * Allocate and initialize an cui item with the given number of extents.
+ */
+struct xfs_cui_log_item *
+xfs_cui_init(
+	struct xfs_mount		*mp,
+	uint				nextents)
+
+{
+	struct xfs_cui_log_item		*cuip;
+
+	ASSERT(nextents > 0);
+	if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
+		cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents),
+				KM_SLEEP);
+	else
+		cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP);
+
+	xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops);
+	cuip->cui_format.cui_nextents = nextents;
+	cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
+	atomic_set(&cuip->cui_next_extent, 0);
+	atomic_set(&cuip->cui_refcount, 2);
+
+	return cuip;
+}
+
+/*
+ * Freeing the CUI requires that we remove it from the AIL if it has already
+ * been placed there. However, the CUI may not yet have been placed in the AIL
+ * when called by xfs_cui_release() from CUD processing due to the ordering of
+ * committed vs unpin operations in bulk insert operations. Hence the reference
+ * count to ensure only the last caller frees the CUI.
+ */
+void
+xfs_cui_release(
+	struct xfs_cui_log_item	*cuip)
+{
+	if (atomic_dec_and_test(&cuip->cui_refcount)) {
+		xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
+		xfs_cui_item_free(cuip);
+	}
+}
+
+static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
+{
+	return container_of(lip, struct xfs_cud_log_item, cud_item);
+}
+
+STATIC void
+xfs_cud_item_size(
+	struct xfs_log_item	*lip,
+	int			*nvecs,
+	int			*nbytes)
+{
+	*nvecs += 1;
+	*nbytes += sizeof(struct xfs_cud_log_format);
+}
+
+/*
+ * This is called to fill in the vector of log iovecs for the
+ * given cud log item. We use only 1 iovec, and we point that
+ * at the cud_log_format structure embedded in the cud item.
+ * It is at this point that we assert that all of the extent
+ * slots in the cud item have been filled.
+ */
+STATIC void
+xfs_cud_item_format(
+	struct xfs_log_item	*lip,
+	struct xfs_log_vec	*lv)
+{
+	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
+
+	cudp->cud_format.cud_type = XFS_LI_CUD;
+	cudp->cud_format.cud_size = 1;
+
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
+			sizeof(struct xfs_cud_log_format));
+}
+
+/*
+ * Pinning has no meaning for an cud item, so just return.
+ */
+STATIC void
+xfs_cud_item_pin(
+	struct xfs_log_item	*lip)
+{
+}
+
+/*
+ * Since pinning has no meaning for an cud item, unpinning does
+ * not either.
+ */
+STATIC void
+xfs_cud_item_unpin(
+	struct xfs_log_item	*lip,
+	int			remove)
+{
+}
+
+/*
+ * There isn't much you can do to push on an cud item.  It is simply stuck
+ * waiting for the log to be flushed to disk.
+ */
+STATIC uint
+xfs_cud_item_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
+{
+	return XFS_ITEM_PINNED;
+}
+
+/*
+ * The CUD is either committed or aborted if the transaction is cancelled. If
+ * the transaction is cancelled, drop our reference to the CUI and free the
+ * CUD.
+ */
+STATIC void
+xfs_cud_item_unlock(
+	struct xfs_log_item	*lip)
+{
+	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
+
+	if (lip->li_flags & XFS_LI_ABORTED) {
+		xfs_cui_release(cudp->cud_cuip);
+		kmem_zone_free(xfs_cud_zone, cudp);
+	}
+}
+
+/*
+ * When the cud item is committed to disk, all we need to do is delete our
+ * reference to our partner cui item and then free ourselves. Since we're
+ * freeing ourselves we must return -1 to keep the transaction code from
+ * further referencing this item.
+ */
+STATIC xfs_lsn_t
+xfs_cud_item_committed(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
+
+	/*
+	 * Drop the CUI reference regardless of whether the CUD has been
+	 * aborted. Once the CUD transaction is constructed, it is the sole
+	 * responsibility of the CUD to release the CUI (even if the CUI is
+	 * aborted due to log I/O error).
+	 */
+	xfs_cui_release(cudp->cud_cuip);
+	kmem_zone_free(xfs_cud_zone, cudp);
+
+	return (xfs_lsn_t)-1;
+}
+
+/*
+ * The CUD dependency tracking op doesn't do squat.  It can't because
+ * it doesn't know where the free extent is coming from.  The dependency
+ * tracking has to be handled by the "enclosing" metadata object.  For
+ * example, for inodes, the inode is locked throughout the extent freeing
+ * so the dependency should be recorded there.
+ */
+STATIC void
+xfs_cud_item_committing(
+	struct xfs_log_item	*lip,
+	xfs_lsn_t		lsn)
+{
+}
+
+/*
+ * This is the ops vector shared by all cud log items.
+ */
+static const struct xfs_item_ops xfs_cud_item_ops = {
+	.iop_size	= xfs_cud_item_size,
+	.iop_format	= xfs_cud_item_format,
+	.iop_pin	= xfs_cud_item_pin,
+	.iop_unpin	= xfs_cud_item_unpin,
+	.iop_unlock	= xfs_cud_item_unlock,
+	.iop_committed	= xfs_cud_item_committed,
+	.iop_push	= xfs_cud_item_push,
+	.iop_committing = xfs_cud_item_committing,
+};
+
+/*
+ * Allocate and initialize an cud item with the given number of extents.
+ */
+struct xfs_cud_log_item *
+xfs_cud_init(
+	struct xfs_mount		*mp,
+	struct xfs_cui_log_item		*cuip)
+
+{
+	struct xfs_cud_log_item	*cudp;
+
+	cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP);
+	xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops);
+	cudp->cud_cuip = cuip;
+	cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
+
+	return cudp;
+}
+
+/*
+ * Process a refcount update intent item that was recovered from the log.
+ * We need to update the refcountbt.
+ */
+int
+xfs_cui_recover(
+	struct xfs_mount		*mp,
+	struct xfs_cui_log_item		*cuip)
+{
+	int				i;
+	int				error = 0;
+	unsigned int			refc_type;
+	struct xfs_phys_extent		*refc;
+	xfs_fsblock_t			startblock_fsb;
+	bool				op_ok;
+	struct xfs_cud_log_item		*cudp;
+	struct xfs_trans		*tp;
+	struct xfs_btree_cur		*rcur = NULL;
+	enum xfs_refcount_intent_type	type;
+	xfs_fsblock_t			firstfsb;
+	xfs_fsblock_t			new_fsb;
+	xfs_extlen_t			new_len;
+	struct xfs_bmbt_irec		irec;
+	struct xfs_defer_ops		dfops;
+	bool				requeue_only = false;
+
+	ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags));
+
+	/*
+	 * First check the validity of the extents described by the
+	 * CUI.  If any are bad, then assume that all are bad and
+	 * just toss the CUI.
+	 */
+	for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
+		refc = &cuip->cui_format.cui_extents[i];
+		startblock_fsb = XFS_BB_TO_FSB(mp,
+				   XFS_FSB_TO_DADDR(mp, refc->pe_startblock));
+		switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
+		case XFS_REFCOUNT_INCREASE:
+		case XFS_REFCOUNT_DECREASE:
+		case XFS_REFCOUNT_ALLOC_COW:
+		case XFS_REFCOUNT_FREE_COW:
+			op_ok = true;
+			break;
+		default:
+			op_ok = false;
+			break;
+		}
+		if (!op_ok || startblock_fsb == 0 ||
+		    refc->pe_len == 0 ||
+		    startblock_fsb >= mp->m_sb.sb_dblocks ||
+		    refc->pe_len >= mp->m_sb.sb_agblocks ||
+		    (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) {
+			/*
+			 * This will pull the CUI from the AIL and
+			 * free the memory associated with it.
+			 */
+			set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
+			xfs_cui_release(cuip);
+			return -EIO;
+		}
+	}
+
+	/*
+	 * Under normal operation, refcount updates are deferred, so we
+	 * wouldn't be adding them directly to a transaction.  All
+	 * refcount updates manage reservation usage internally and
+	 * dynamically by deferring work that won't fit in the
+	 * transaction.  Normally, any work that needs to be deferred
+	 * gets attached to the same defer_ops that scheduled the
+	 * refcount update.  However, we're in log recovery here, so we
+	 * we create our own defer_ops and use that to finish up any
+	 * work that doesn't fit.
+	 */
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
+	if (error)
+		return error;
+	cudp = xfs_trans_get_cud(tp, cuip);
+
+	xfs_defer_init(&dfops, &firstfsb);
+	for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
+		refc = &cuip->cui_format.cui_extents[i];
+		refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
+		switch (refc_type) {
+		case XFS_REFCOUNT_INCREASE:
+		case XFS_REFCOUNT_DECREASE:
+		case XFS_REFCOUNT_ALLOC_COW:
+		case XFS_REFCOUNT_FREE_COW:
+			type = refc_type;
+			break;
+		default:
+			error = -EFSCORRUPTED;
+			goto abort_error;
+		}
+		if (requeue_only) {
+			new_fsb = refc->pe_startblock;
+			new_len = refc->pe_len;
+		} else
+			error = xfs_trans_log_finish_refcount_update(tp, cudp,
+				&dfops, type, refc->pe_startblock, refc->pe_len,
+				&new_fsb, &new_len, &rcur);
+		if (error)
+			goto abort_error;
+
+		/* Requeue what we didn't finish. */
+		if (new_len > 0) {
+			irec.br_startblock = new_fsb;
+			irec.br_blockcount = new_len;
+			switch (type) {
+			case XFS_REFCOUNT_INCREASE:
+				error = xfs_refcount_increase_extent(
+						tp->t_mountp, &dfops, &irec);
+				break;
+			case XFS_REFCOUNT_DECREASE:
+				error = xfs_refcount_decrease_extent(
+						tp->t_mountp, &dfops, &irec);
+				break;
+			case XFS_REFCOUNT_ALLOC_COW:
+				error = xfs_refcount_alloc_cow_extent(
+						tp->t_mountp, &dfops,
+						irec.br_startblock,
+						irec.br_blockcount);
+				break;
+			case XFS_REFCOUNT_FREE_COW:
+				error = xfs_refcount_free_cow_extent(
+						tp->t_mountp, &dfops,
+						irec.br_startblock,
+						irec.br_blockcount);
+				break;
+			default:
+				ASSERT(0);
+			}
+			if (error)
+				goto abort_error;
+			requeue_only = true;
+		}
+	}
+
+	xfs_refcount_finish_one_cleanup(tp, rcur, error);
+	error = xfs_defer_finish(&tp, &dfops, NULL);
+	if (error)
+		goto abort_error;
+	set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
+	error = xfs_trans_commit(tp);
+	return error;
+
+abort_error:
+	xfs_refcount_finish_one_cleanup(tp, rcur, error);
+	xfs_defer_cancel(&dfops);
+	xfs_trans_cancel(tp);
+	return error;
+}

fs/xfs/xfs_refcount_item.h

@@ -0,0 +1,101 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef	__XFS_REFCOUNT_ITEM_H__
+#define	__XFS_REFCOUNT_ITEM_H__
+
+/*
+ * There are (currently) two pairs of refcount btree redo item types:
+ * increase and decrease.  The log items for these are CUI (refcount
+ * update intent) and CUD (refcount update done).  The redo item type
+ * is encoded in the flags field of each xfs_map_extent.
+ *
+ * *I items should be recorded in the *first* of a series of rolled
+ * transactions, and the *D items should be recorded in the same
+ * transaction that records the associated refcountbt updates.
+ *
+ * Should the system crash after the commit of the first transaction
+ * but before the commit of the final transaction in a series, log
+ * recovery will use the redo information recorded by the intent items
+ * to replay the refcountbt metadata updates.
+ */
+
+/* kernel only CUI/CUD definitions */
+
+struct xfs_mount;
+struct kmem_zone;
+
+/*
+ * Max number of extents in fast allocation path.
+ */
+#define	XFS_CUI_MAX_FAST_EXTENTS	16
+
+/*
+ * Define CUI flag bits. Manipulated by set/clear/test_bit operators.
+ */
+#define	XFS_CUI_RECOVERED		1
+
+/*
+ * This is the "refcount update intent" log item.  It is used to log
+ * the fact that some reverse mappings need to change.  It is used in
+ * conjunction with the "refcount update done" log item described
+ * below.
+ *
+ * These log items follow the same rules as struct xfs_efi_log_item;
+ * see the comments about that structure (in xfs_extfree_item.h) for
+ * more details.
+ */
+struct xfs_cui_log_item {
+	struct xfs_log_item		cui_item;
+	atomic_t			cui_refcount;
+	atomic_t			cui_next_extent;
+	unsigned long			cui_flags;	/* misc flags */
+	struct xfs_cui_log_format	cui_format;
+};
+
+static inline size_t
+xfs_cui_log_item_sizeof(
+	unsigned int		nr)
+{
+	return offsetof(struct xfs_cui_log_item, cui_format) +
+			xfs_cui_log_format_sizeof(nr);
+}
+
+/*
+ * This is the "refcount update done" log item.  It is used to log the
+ * fact that some refcountbt updates mentioned in an earlier cui item
+ * have been performed.
+ */
+struct xfs_cud_log_item {
+	struct xfs_log_item		cud_item;
+	struct xfs_cui_log_item		*cud_cuip;
+	struct xfs_cud_log_format	cud_format;
+};
+
+extern struct kmem_zone	*xfs_cui_zone;
+extern struct kmem_zone	*xfs_cud_zone;
+
+struct xfs_cui_log_item *xfs_cui_init(struct xfs_mount *, uint);
+struct xfs_cud_log_item *xfs_cud_init(struct xfs_mount *,
+		struct xfs_cui_log_item *);
+void xfs_cui_item_free(struct xfs_cui_log_item *);
+void xfs_cui_release(struct xfs_cui_log_item *);
+int xfs_cui_recover(struct xfs_mount *mp, struct xfs_cui_log_item *cuip);
+
+#endif	/* __XFS_REFCOUNT_ITEM_H__ */

fs/xfs/xfs_reflink.c

@@ -0,0 +1,1688 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#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"
+#include "xfs_bmap_btree.h"
+#include "xfs_trans_space.h"
+#include "xfs_bit.h"
+#include "xfs_alloc.h"
+#include "xfs_quota_defs.h"
+#include "xfs_quota.h"
+#include "xfs_btree.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_reflink.h"
+#include "xfs_iomap.h"
+#include "xfs_rmap_btree.h"
+#include "xfs_sb.h"
+#include "xfs_ag_resv.h"
+
+/*
+ * Copy on Write of Shared Blocks
+ *
+ * XFS must preserve "the usual" file semantics even when two files share
+ * the same physical blocks.  This means that a write to one file must not
+ * alter the blocks in a different file; the way that we'll do that is
+ * through the use of a copy-on-write mechanism.  At a high level, that
+ * means that when we want to write to a shared block, we allocate a new
+ * block, write the data to the new block, and if that succeeds we map the
+ * new block into the file.
+ *
+ * XFS provides a "delayed allocation" mechanism that defers the allocation
+ * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
+ * possible.  This reduces fragmentation by enabling the filesystem to ask
+ * for bigger chunks less often, which is exactly what we want for CoW.
+ *
+ * The delalloc mechanism begins when the kernel wants to make a block
+ * writable (write_begin or page_mkwrite).  If the offset is not mapped, we
+ * create a delalloc mapping, which is a regular in-core extent, but without
+ * a real startblock.  (For delalloc mappings, the startblock encodes both
+ * a flag that this is a delalloc mapping, and a worst-case estimate of how
+ * many blocks might be required to put the mapping into the BMBT.)  delalloc
+ * mappings are a reservation against the free space in the filesystem;
+ * adjacent mappings can also be combined into fewer larger mappings.
+ *
+ * When dirty pages are being written out (typically in writepage), the
+ * delalloc reservations are converted into real mappings by allocating
+ * blocks and replacing the delalloc mapping with real ones.  A delalloc
+ * mapping can be replaced by several real ones if the free space is
+ * fragmented.
+ *
+ * We want to adapt the delalloc mechanism for copy-on-write, since the
+ * write paths are similar.  The first two steps (creating the reservation
+ * and allocating the blocks) are exactly the same as delalloc except that
+ * the mappings must be stored in a separate CoW fork because we do not want
+ * to disturb the mapping in the data fork until we're sure that the write
+ * succeeded.  IO completion in this case is the process of removing the old
+ * mapping from the data fork and moving the new mapping from the CoW fork to
+ * the data fork.  This will be discussed shortly.
+ *
+ * For now, unaligned directio writes will be bounced back to the page cache.
+ * Block-aligned directio writes will use the same mechanism as buffered
+ * writes.
+ *
+ * CoW remapping must be done after the data block write completes,
+ * because we don't want to destroy the old data fork map until we're sure
+ * the new block has been written.  Since the new mappings are kept in a
+ * separate fork, we can simply iterate these mappings to find the ones
+ * that cover the file blocks that we just CoW'd.  For each extent, simply
+ * unmap the corresponding range in the data fork, map the new range into
+ * the data fork, and remove the extent from the CoW fork.
+ *
+ * Since the remapping operation can be applied to an arbitrary file
+ * range, we record the need for the remap step as a flag in the ioend
+ * instead of declaring a new IO type.  This is required for direct io
+ * because we only have ioend for the whole dio, and we have to be able to
+ * remember the presence of unwritten blocks and CoW blocks with a single
+ * ioend structure.  Better yet, the more ground we can cover with one
+ * ioend, the better.
+ */
+
+/*
+ * 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.
+ */
+int
+xfs_reflink_find_shared(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_agblock_t		agbno,
+	xfs_extlen_t		aglen,
+	xfs_agblock_t		*fbno,
+	xfs_extlen_t		*flen,
+	bool			find_end_of_shared)
+{
+	struct xfs_buf		*agbp;
+	struct xfs_btree_cur	*cur;
+	int			error;
+
+	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
+	if (error)
+		return error;
+
+	cur = xfs_refcountbt_init_cursor(mp, NULL, agbp, agno, NULL);
+
+	error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
+			find_end_of_shared);
+
+	xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
+
+	xfs_buf_relse(agbp);
+	return error;
+}
+
+/*
+ * Trim the mapping to the next block where there's a change in the
+ * shared/unshared status.  More specifically, this means that we
+ * find the lowest-numbered extent of shared blocks that coincides with
+ * the given block mapping.  If the shared extent overlaps the start of
+ * the mapping, trim the mapping to the end of the shared extent.  If
+ * the shared region intersects the mapping, trim the mapping to the
+ * start of the shared extent.  If there are no shared regions that
+ * overlap, just return the original extent.
+ */
+int
+xfs_reflink_trim_around_shared(
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*irec,
+	bool			*shared,
+	bool			*trimmed)
+{
+	xfs_agnumber_t		agno;
+	xfs_agblock_t		agbno;
+	xfs_extlen_t		aglen;
+	xfs_agblock_t		fbno;
+	xfs_extlen_t		flen;
+	int			error = 0;
+
+	/* Holes, unwritten, and delalloc extents cannot be shared */
+	if (!xfs_is_reflink_inode(ip) ||
+	    ISUNWRITTEN(irec) ||
+	    irec->br_startblock == HOLESTARTBLOCK ||
+	    irec->br_startblock == DELAYSTARTBLOCK) {
+		*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, agno, agbno,
+			aglen, &fbno, &flen, true);
+	if (error)
+		return error;
+
+	*shared = *trimmed = false;
+	if (fbno == NULLAGBLOCK) {
+		/* No shared blocks at all. */
+		return 0;
+	} else if (fbno == agbno) {
+		/*
+		 * 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.
+		 */
+		irec->br_blockcount = flen;
+		*shared = true;
+		if (flen != aglen)
+			*trimmed = 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
+		 * start of the shared region.
+		 */
+		irec->br_blockcount = fbno - agbno;
+		*trimmed = true;
+		return 0;
+	}
+}
+
+/* Create a CoW reservation for a range of blocks within a file. */
+static int
+__xfs_reflink_reserve_cow(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		*offset_fsb,
+	xfs_fileoff_t		end_fsb,
+	bool			*skipped)
+{
+	struct xfs_bmbt_irec	got, prev, imap;
+	xfs_fileoff_t		orig_end_fsb;
+	int			nimaps, eof = 0, error = 0;
+	bool			shared = false, trimmed = false;
+	xfs_extnum_t		idx;
+	xfs_extlen_t		align;
+
+	/* Already reserved?  Skip the refcount btree access. */
+	xfs_bmap_search_extents(ip, *offset_fsb, XFS_COW_FORK, &eof, &idx,
+			&got, &prev);
+	if (!eof && got.br_startoff <= *offset_fsb) {
+		end_fsb = orig_end_fsb = got.br_startoff + got.br_blockcount;
+		trace_xfs_reflink_cow_found(ip, &got);
+		goto done;
+	}
+
+	/* Read extent from the source file. */
+	nimaps = 1;
+	error = xfs_bmapi_read(ip, *offset_fsb, end_fsb - *offset_fsb,
+			&imap, &nimaps, 0);
+	if (error)
+		goto out_unlock;
+	ASSERT(nimaps == 1);
+
+	/* Trim the mapping to the nearest shared extent boundary. */
+	error = xfs_reflink_trim_around_shared(ip, &imap, &shared, &trimmed);
+	if (error)
+		goto out_unlock;
+
+	end_fsb = orig_end_fsb = imap.br_startoff + imap.br_blockcount;
+
+	/* Not shared?  Just report the (potentially capped) extent. */
+	if (!shared) {
+		*skipped = true;
+		goto done;
+	}
+
+	/*
+	 * Fork all the shared blocks from our write offset until the end of
+	 * the extent.
+	 */
+	error = xfs_qm_dqattach_locked(ip, 0);
+	if (error)
+		goto out_unlock;
+
+	align = xfs_eof_alignment(ip, xfs_get_cowextsz_hint(ip));
+	if (align)
+		end_fsb = roundup_64(end_fsb, align);
+
+retry:
+	error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, *offset_fsb,
+			end_fsb - *offset_fsb, &got,
+			&prev, &idx, eof);
+	switch (error) {
+	case 0:
+		break;
+	case -ENOSPC:
+	case -EDQUOT:
+		/* retry without any preallocation */
+		trace_xfs_reflink_cow_enospc(ip, &imap);
+		if (end_fsb != orig_end_fsb) {
+			end_fsb = orig_end_fsb;
+			goto retry;
+		}
+		/*FALLTHRU*/
+	default:
+		goto out_unlock;
+	}
+
+	if (end_fsb != orig_end_fsb)
+		xfs_inode_set_cowblocks_tag(ip);
+
+	trace_xfs_reflink_cow_alloc(ip, &got);
+done:
+	*offset_fsb = end_fsb;
+out_unlock:
+	return error;
+}
+
+/* Create a CoW reservation for part of a file. */
+int
+xfs_reflink_reserve_cow_range(
+	struct xfs_inode	*ip,
+	xfs_off_t		offset,
+	xfs_off_t		count)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	xfs_fileoff_t		offset_fsb, end_fsb;
+	bool			skipped = false;
+	int			error;
+
+	trace_xfs_reflink_reserve_cow_range(ip, offset, count);
+
+	offset_fsb = XFS_B_TO_FSBT(mp, offset);
+	end_fsb = XFS_B_TO_FSB(mp, offset + count);
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	while (offset_fsb < end_fsb) {
+		error = __xfs_reflink_reserve_cow(ip, &offset_fsb, end_fsb,
+				&skipped);
+		if (error) {
+			trace_xfs_reflink_reserve_cow_range_error(ip, error,
+				_RET_IP_);
+			break;
+		}
+	}
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+
+	return error;
+}
+
+/* Allocate all CoW reservations covering a range of blocks in a file. */
+static int
+__xfs_reflink_allocate_cow(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		*offset_fsb,
+	xfs_fileoff_t		end_fsb)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_bmbt_irec	imap;
+	struct xfs_defer_ops	dfops;
+	struct xfs_trans	*tp;
+	xfs_fsblock_t		first_block;
+	xfs_fileoff_t		next_fsb;
+	int			nimaps = 1, error;
+	bool			skipped = false;
+
+	xfs_defer_init(&dfops, &first_block);
+
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
+			XFS_TRANS_RESERVE, &tp);
+	if (error)
+		return error;
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+
+	next_fsb = *offset_fsb;
+	error = __xfs_reflink_reserve_cow(ip, &next_fsb, end_fsb, &skipped);
+	if (error)
+		goto out_trans_cancel;
+
+	if (skipped) {
+		*offset_fsb = next_fsb;
+		goto out_trans_cancel;
+	}
+
+	xfs_trans_ijoin(tp, ip, 0);
+	error = xfs_bmapi_write(tp, ip, *offset_fsb, next_fsb - *offset_fsb,
+			XFS_BMAPI_COWFORK, &first_block,
+			XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK),
+			&imap, &nimaps, &dfops);
+	if (error)
+		goto out_trans_cancel;
+
+	/* We might not have been able to map the whole delalloc extent */
+	*offset_fsb = min(*offset_fsb + imap.br_blockcount, next_fsb);
+
+	error = xfs_defer_finish(&tp, &dfops, NULL);
+	if (error)
+		goto out_trans_cancel;
+
+	error = xfs_trans_commit(tp);
+
+out_unlock:
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	return error;
+out_trans_cancel:
+	xfs_defer_cancel(&dfops);
+	xfs_trans_cancel(tp);
+	goto out_unlock;
+}
+
+/* Allocate all CoW reservations covering a part of a file. */
+int
+xfs_reflink_allocate_cow_range(
+	struct xfs_inode	*ip,
+	xfs_off_t		offset,
+	xfs_off_t		count)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
+	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
+	int			error;
+
+	ASSERT(xfs_is_reflink_inode(ip));
+
+	trace_xfs_reflink_allocate_cow_range(ip, offset, count);
+
+	/*
+	 * Make sure that the dquots are there.
+	 */
+	error = xfs_qm_dqattach(ip, 0);
+	if (error)
+		return error;
+
+	while (offset_fsb < end_fsb) {
+		error = __xfs_reflink_allocate_cow(ip, &offset_fsb, end_fsb);
+		if (error) {
+			trace_xfs_reflink_allocate_cow_range_error(ip, error,
+					_RET_IP_);
+			break;
+		}
+	}
+
+	return error;
+}
+
+/*
+ * Find the CoW reservation (and whether or not it needs block allocation)
+ * for a given byte offset of a file.
+ */
+bool
+xfs_reflink_find_cow_mapping(
+	struct xfs_inode		*ip,
+	xfs_off_t			offset,
+	struct xfs_bmbt_irec		*imap,
+	bool				*need_alloc)
+{
+	struct xfs_bmbt_irec		irec;
+	struct xfs_ifork		*ifp;
+	struct xfs_bmbt_rec_host	*gotp;
+	xfs_fileoff_t			bno;
+	xfs_extnum_t			idx;
+
+	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
+	ASSERT(xfs_is_reflink_inode(ip));
+
+	/* Find the extent in the CoW fork. */
+	ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
+	bno = XFS_B_TO_FSBT(ip->i_mount, offset);
+	gotp = xfs_iext_bno_to_ext(ifp, bno, &idx);
+	if (!gotp)
+		return false;
+
+	xfs_bmbt_get_all(gotp, &irec);
+	if (bno >= irec.br_startoff + irec.br_blockcount ||
+	    bno < irec.br_startoff)
+		return false;
+
+	trace_xfs_reflink_find_cow_mapping(ip, offset, 1, XFS_IO_OVERWRITE,
+			&irec);
+
+	/* If it's still delalloc, we must allocate later. */
+	*imap = irec;
+	*need_alloc = !!(isnullstartblock(irec.br_startblock));
+
+	return true;
+}
+
+/*
+ * Trim an extent to end at the next CoW reservation past offset_fsb.
+ */
+int
+xfs_reflink_trim_irec_to_next_cow(
+	struct xfs_inode		*ip,
+	xfs_fileoff_t			offset_fsb,
+	struct xfs_bmbt_irec		*imap)
+{
+	struct xfs_bmbt_irec		irec;
+	struct xfs_ifork		*ifp;
+	struct xfs_bmbt_rec_host	*gotp;
+	xfs_extnum_t			idx;
+
+	if (!xfs_is_reflink_inode(ip))
+		return 0;
+
+	/* Find the extent in the CoW fork. */
+	ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
+	gotp = xfs_iext_bno_to_ext(ifp, offset_fsb, &idx);
+	if (!gotp)
+		return 0;
+	xfs_bmbt_get_all(gotp, &irec);
+
+	/* This is the extent before; try sliding up one. */
+	if (irec.br_startoff < offset_fsb) {
+		idx++;
+		if (idx >= ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
+			return 0;
+		gotp = xfs_iext_get_ext(ifp, idx);
+		xfs_bmbt_get_all(gotp, &irec);
+	}
+
+	if (irec.br_startoff >= imap->br_startoff + imap->br_blockcount)
+		return 0;
+
+	imap->br_blockcount = irec.br_startoff - imap->br_startoff;
+	trace_xfs_reflink_trim_irec(ip, imap);
+
+	return 0;
+}
+
+/*
+ * Cancel all pending CoW reservations for some block range of an inode.
+ */
+int
+xfs_reflink_cancel_cow_blocks(
+	struct xfs_inode		*ip,
+	struct xfs_trans		**tpp,
+	xfs_fileoff_t			offset_fsb,
+	xfs_fileoff_t			end_fsb)
+{
+	struct xfs_bmbt_irec		irec;
+	xfs_filblks_t			count_fsb;
+	xfs_fsblock_t			firstfsb;
+	struct xfs_defer_ops		dfops;
+	int				error = 0;
+	int				nimaps;
+
+	if (!xfs_is_reflink_inode(ip))
+		return 0;
+
+	/* Go find the old extent in the CoW fork. */
+	while (offset_fsb < end_fsb) {
+		nimaps = 1;
+		count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
+		error = xfs_bmapi_read(ip, offset_fsb, count_fsb, &irec,
+				&nimaps, XFS_BMAPI_COWFORK);
+		if (error)
+			break;
+		ASSERT(nimaps == 1);
+
+		trace_xfs_reflink_cancel_cow(ip, &irec);
+
+		if (irec.br_startblock == DELAYSTARTBLOCK) {
+			/* Free a delayed allocation. */
+			xfs_mod_fdblocks(ip->i_mount, irec.br_blockcount,
+					false);
+			ip->i_delayed_blks -= irec.br_blockcount;
+
+			/* Remove the mapping from the CoW fork. */
+			error = xfs_bunmapi_cow(ip, &irec);
+			if (error)
+				break;
+		} else if (irec.br_startblock == HOLESTARTBLOCK) {
+			/* empty */
+		} else {
+			xfs_trans_ijoin(*tpp, ip, 0);
+			xfs_defer_init(&dfops, &firstfsb);
+
+			/* Free the CoW orphan record. */
+			error = xfs_refcount_free_cow_extent(ip->i_mount,
+					&dfops, irec.br_startblock,
+					irec.br_blockcount);
+			if (error)
+				break;
+
+			xfs_bmap_add_free(ip->i_mount, &dfops,
+					irec.br_startblock, irec.br_blockcount,
+					NULL);
+
+			/* Update quota accounting */
+			xfs_trans_mod_dquot_byino(*tpp, ip, XFS_TRANS_DQ_BCOUNT,
+					-(long)irec.br_blockcount);
+
+			/* Roll the transaction */
+			error = xfs_defer_finish(tpp, &dfops, ip);
+			if (error) {
+				xfs_defer_cancel(&dfops);
+				break;
+			}
+
+			/* Remove the mapping from the CoW fork. */
+			error = xfs_bunmapi_cow(ip, &irec);
+			if (error)
+				break;
+		}
+
+		/* Roll on... */
+		offset_fsb = irec.br_startoff + irec.br_blockcount;
+	}
+
+	return error;
+}
+
+/*
+ * Cancel all pending CoW reservations for some byte range of an inode.
+ */
+int
+xfs_reflink_cancel_cow_range(
+	struct xfs_inode	*ip,
+	xfs_off_t		offset,
+	xfs_off_t		count)
+{
+	struct xfs_trans	*tp;
+	xfs_fileoff_t		offset_fsb;
+	xfs_fileoff_t		end_fsb;
+	int			error;
+
+	trace_xfs_reflink_cancel_cow_range(ip, offset, count);
+	ASSERT(xfs_is_reflink_inode(ip));
+
+	offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
+	if (count == NULLFILEOFF)
+		end_fsb = NULLFILEOFF;
+	else
+		end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
+
+	/* Start a rolling transaction to remove the mappings */
+	error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
+			0, 0, 0, &tp);
+	if (error)
+		goto out;
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, ip, 0);
+
+	/* Scrape out the old CoW reservations */
+	error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb);
+	if (error)
+		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);
+out:
+	trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Remap parts of a file's data fork after a successful CoW.
+ */
+int
+xfs_reflink_end_cow(
+	struct xfs_inode		*ip,
+	xfs_off_t			offset,
+	xfs_off_t			count)
+{
+	struct xfs_bmbt_irec		irec;
+	struct xfs_bmbt_irec		uirec;
+	struct xfs_trans		*tp;
+	xfs_fileoff_t			offset_fsb;
+	xfs_fileoff_t			end_fsb;
+	xfs_filblks_t			count_fsb;
+	xfs_fsblock_t			firstfsb;
+	struct xfs_defer_ops		dfops;
+	int				error;
+	unsigned int			resblks;
+	xfs_filblks_t			ilen;
+	xfs_filblks_t			rlen;
+	int				nimaps;
+
+	trace_xfs_reflink_end_cow(ip, offset, count);
+
+	offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
+	end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
+	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
+
+	/* Start a rolling transaction to switch the mappings */
+	resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
+	error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
+			resblks, 0, 0, &tp);
+	if (error)
+		goto out;
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, ip, 0);
+
+	/* Go find the old extent in the CoW fork. */
+	while (offset_fsb < end_fsb) {
+		/* Read extent from the source file */
+		nimaps = 1;
+		count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
+		error = xfs_bmapi_read(ip, offset_fsb, count_fsb, &irec,
+				&nimaps, XFS_BMAPI_COWFORK);
+		if (error)
+			goto out_cancel;
+		ASSERT(nimaps == 1);
+
+		ASSERT(irec.br_startblock != DELAYSTARTBLOCK);
+		trace_xfs_reflink_cow_remap(ip, &irec);
+
+		/*
+		 * We can have a hole in the CoW fork if part of a directio
+		 * write is CoW but part of it isn't.
+		 */
+		rlen = ilen = irec.br_blockcount;
+		if (irec.br_startblock == HOLESTARTBLOCK)
+			goto next_extent;
+
+		/* Unmap the old blocks in the data fork. */
+		while (rlen) {
+			xfs_defer_init(&dfops, &firstfsb);
+			error = __xfs_bunmapi(tp, ip, irec.br_startoff,
+					&rlen, 0, 1, &firstfsb, &dfops);
+			if (error)
+				goto out_defer;
+
+			/*
+			 * 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 = irec.br_blockcount - rlen;
+			irec.br_blockcount = rlen;
+			trace_xfs_reflink_cow_remap_piece(ip, &uirec);
+
+			/* Free the CoW orphan record. */
+			error = xfs_refcount_free_cow_extent(tp->t_mountp,
+					&dfops, uirec.br_startblock,
+					uirec.br_blockcount);
+			if (error)
+				goto out_defer;
+
+			/* Map the new blocks into the data fork. */
+			error = xfs_bmap_map_extent(tp->t_mountp, &dfops,
+					ip, &uirec);
+			if (error)
+				goto out_defer;
+
+			/* Remove the mapping from the CoW fork. */
+			error = xfs_bunmapi_cow(ip, &uirec);
+			if (error)
+				goto out_defer;
+
+			error = xfs_defer_finish(&tp, &dfops, ip);
+			if (error)
+				goto out_defer;
+		}
+
+next_extent:
+		/* Roll on... */
+		offset_fsb = irec.br_startoff + ilen;
+	}
+
+	error = xfs_trans_commit(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	if (error)
+		goto out;
+	return 0;
+
+out_defer:
+	xfs_defer_cancel(&dfops);
+out_cancel:
+	xfs_trans_cancel(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+out:
+	trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Free leftover CoW reservations that didn't get cleaned out.
+ */
+int
+xfs_reflink_recover_cow(
+	struct xfs_mount	*mp)
+{
+	xfs_agnumber_t		agno;
+	int			error = 0;
+
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return 0;
+
+	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
+		error = xfs_refcount_recover_cow_leftovers(mp, agno);
+		if (error)
+			break;
+	}
+
+	return error;
+}
+
+/*
+ * Reflinking (Block) Ranges of Two Files Together
+ *
+ * First, ensure that the reflink flag is set on both inodes.  The flag is an
+ * optimization to avoid unnecessary refcount btree lookups in the write path.
+ *
+ * Now we can iteratively remap the range of extents (and holes) in src to the
+ * corresponding ranges in dest.  Let drange and srange denote the ranges of
+ * logical blocks in dest and src touched by the reflink operation.
+ *
+ * While the length of drange is greater than zero,
+ *    - Read src's bmbt at the start of srange ("imap")
+ *    - If imap doesn't exist, make imap appear to start at the end of srange
+ *      with zero length.
+ *    - If imap starts before srange, advance imap to start at srange.
+ *    - If imap goes beyond srange, truncate imap to end at the end of srange.
+ *    - Punch (imap start - srange start + imap len) blocks from dest at
+ *      offset (drange start).
+ *    - If imap points to a real range of pblks,
+ *         > Increase the refcount of the imap's pblks
+ *         > Map imap's pblks into dest at the offset
+ *           (drange start + imap start - srange start)
+ *    - Advance drange and srange by (imap start - srange start + imap len)
+ *
+ * Finally, if the reflink made dest longer, update both the in-core and
+ * on-disk file sizes.
+ *
+ * ASCII Art Demonstration:
+ *
+ * Let's say we want to reflink this source file:
+ *
+ * ----SSSSSSS-SSSSS----SSSSSS (src file)
+ *   <-------------------->
+ *
+ * into this destination file:
+ *
+ * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
+ *        <-------------------->
+ * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
+ * Observe that the range has different logical offsets in either file.
+ *
+ * Consider that the first extent in the source file doesn't line up with our
+ * reflink range.  Unmapping  and remapping are separate operations, so we can
+ * unmap more blocks from the destination file than we remap.
+ *
+ * ----SSSSSSS-SSSSS----SSSSSS
+ *   <------->
+ * --DDDDD---------DDDDD--DDD
+ *        <------->
+ *
+ * Now remap the source extent into the destination file:
+ *
+ * ----SSSSSSS-SSSSS----SSSSSS
+ *   <------->
+ * --DDDDD--SSSSSSSDDDDD--DDD
+ *        <------->
+ *
+ * Do likewise with the second hole and extent in our range.  Holes in the
+ * unmap range don't affect our operation.
+ *
+ * ----SSSSSSS-SSSSS----SSSSSS
+ *            <---->
+ * --DDDDD--SSSSSSS-SSSSS-DDD
+ *                 <---->
+ *
+ * Finally, unmap and remap part of the third extent.  This will increase the
+ * size of the destination file.
+ *
+ * ----SSSSSSS-SSSSS----SSSSSS
+ *                  <----->
+ * --DDDDD--SSSSSSS-SSSSS----SSS
+ *                       <----->
+ *
+ * Once we update the destination file's i_size, we're done.
+ */
+
+/*
+ * Ensure the reflink bit is set in both inodes.
+ */
+STATIC int
+xfs_reflink_set_inode_flag(
+	struct xfs_inode	*src,
+	struct xfs_inode	*dest)
+{
+	struct xfs_mount	*mp = src->i_mount;
+	int			error;
+	struct xfs_trans	*tp;
+
+	if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
+		return 0;
+
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
+	if (error)
+		goto out_error;
+
+	/* Lock both files against IO */
+	if (src->i_ino == dest->i_ino)
+		xfs_ilock(src, XFS_ILOCK_EXCL);
+	else
+		xfs_lock_two_inodes(src, dest, XFS_ILOCK_EXCL);
+
+	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;
+		xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
+		xfs_ifork_init_cow(src);
+	} else
+		xfs_iunlock(src, XFS_ILOCK_EXCL);
+
+	if (src->i_ino == dest->i_ino)
+		goto commit_flags;
+
+	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;
+		xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
+		xfs_ifork_init_cow(dest);
+	} else
+		xfs_iunlock(dest, XFS_ILOCK_EXCL);
+
+commit_flags:
+	error = xfs_trans_commit(tp);
+	if (error)
+		goto out_error;
+	return error;
+
+out_error:
+	trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Update destination inode size & cowextsize hint, if necessary.
+ */
+STATIC int
+xfs_reflink_update_dest(
+	struct xfs_inode	*dest,
+	xfs_off_t		newlen,
+	xfs_extlen_t		cowextsize)
+{
+	struct xfs_mount	*mp = dest->i_mount;
+	struct xfs_trans	*tp;
+	int			error;
+
+	if (newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0)
+		return 0;
+
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
+	if (error)
+		goto out_error;
+
+	xfs_ilock(dest, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
+
+	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;
+	}
+
+	if (cowextsize) {
+		dest->i_d.di_cowextsize = cowextsize;
+		dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
+	}
+
+	xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
+
+	error = xfs_trans_commit(tp);
+	if (error)
+		goto out_error;
+	return error;
+
+out_error:
+	trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Do we have enough reserve in this AG to handle a reflink?  The refcount
+ * btree already reserved all the space it needs, but the rmap btree can grow
+ * infinitely, so we won't allow more reflinks when the AG is down to the
+ * btree reserves.
+ */
+static int
+xfs_reflink_ag_has_free_space(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno)
+{
+	struct xfs_perag	*pag;
+	int			error = 0;
+
+	if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
+		return 0;
+
+	pag = xfs_perag_get(mp, agno);
+	if (xfs_ag_resv_critical(pag, XFS_AG_RESV_AGFL) ||
+	    xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
+		error = -ENOSPC;
+	xfs_perag_put(pag);
+	return error;
+}
+
+/*
+ * 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.
+ */
+STATIC int
+xfs_reflink_remap_extent(
+	struct xfs_inode	*ip,
+	struct xfs_bmbt_irec	*irec,
+	xfs_fileoff_t		destoff,
+	xfs_off_t		new_isize)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	xfs_fsblock_t		firstfsb;
+	unsigned int		resblks;
+	struct xfs_defer_ops	dfops;
+	struct xfs_bmbt_irec	uirec;
+	bool			real_extent;
+	xfs_filblks_t		rlen;
+	xfs_filblks_t		unmap_len;
+	xfs_off_t		newlen;
+	int			error;
+
+	unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
+	trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
+
+	/* Only remap normal extents. */
+	real_extent =  (irec->br_startblock != HOLESTARTBLOCK &&
+			irec->br_startblock != DELAYSTARTBLOCK &&
+			!ISUNWRITTEN(irec));
+
+	/* 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 */
+	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);
+
+	/* 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 (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) {
+		xfs_defer_init(&dfops, &firstfsb);
+		error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1,
+				&firstfsb, &dfops);
+		if (error)
+			goto out_defer;
+
+		/*
+		 * 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 this isn't a real mapping, we're done. */
+		if (!real_extent || uirec.br_blockcount == 0)
+			goto next_extent;
+
+		trace_xfs_reflink_remap(ip, uirec.br_startoff,
+				uirec.br_blockcount, uirec.br_startblock);
+
+		/* Update the refcount tree */
+		error = xfs_refcount_increase_extent(mp, &dfops, &uirec);
+		if (error)
+			goto out_defer;
+
+		/* Map the new blocks into the data fork. */
+		error = xfs_bmap_map_extent(mp, &dfops, ip, &uirec);
+		if (error)
+			goto out_defer;
+
+		/* 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);
+		}
+
+next_extent:
+		/* Process all the deferred stuff. */
+		error = xfs_defer_finish(&tp, &dfops, ip);
+		if (error)
+			goto out_defer;
+	}
+
+	error = xfs_trans_commit(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	if (error)
+		goto out;
+	return 0;
+
+out_defer:
+	xfs_defer_cancel(&dfops);
+out_cancel:
+	xfs_trans_cancel(tp);
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+out:
+	trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Iteratively remap one file's extents (and holes) to another's.
+ */
+STATIC int
+xfs_reflink_remap_blocks(
+	struct xfs_inode	*src,
+	xfs_fileoff_t		srcoff,
+	struct xfs_inode	*dest,
+	xfs_fileoff_t		destoff,
+	xfs_filblks_t		len,
+	xfs_off_t		new_isize)
+{
+	struct xfs_bmbt_irec	imap;
+	int			nimaps;
+	int			error = 0;
+	xfs_filblks_t		range_len;
+
+	/* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
+	while (len) {
+		trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
+				dest, destoff);
+		/* Read extent from the source file */
+		nimaps = 1;
+		xfs_ilock(src, XFS_ILOCK_EXCL);
+		error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
+		xfs_iunlock(src, XFS_ILOCK_EXCL);
+		if (error)
+			goto err;
+		ASSERT(nimaps == 1);
+
+		trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
+				&imap);
+
+		/* Translate imap into the destination file. */
+		range_len = imap.br_startoff + imap.br_blockcount - srcoff;
+		imap.br_startoff += destoff - srcoff;
+
+		/* Clear dest from destoff to the end of imap and map it in. */
+		error = xfs_reflink_remap_extent(dest, &imap, destoff,
+				new_isize);
+		if (error)
+			goto err;
+
+		if (fatal_signal_pending(current)) {
+			error = -EINTR;
+			goto err;
+		}
+
+		/* Advance drange/srange */
+		srcoff += range_len;
+		destoff += range_len;
+		len -= range_len;
+	}
+
+	return 0;
+
+err:
+	trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Read a page's worth of file data into the page cache.  Return the page
+ * locked.
+ */
+static struct page *
+xfs_get_page(
+	struct inode	*inode,
+	xfs_off_t	offset)
+{
+	struct address_space	*mapping;
+	struct page		*page;
+	pgoff_t			n;
+
+	n = offset >> PAGE_SHIFT;
+	mapping = inode->i_mapping;
+	page = read_mapping_page(mapping, n, NULL);
+	if (IS_ERR(page))
+		return page;
+	if (!PageUptodate(page)) {
+		put_page(page);
+		return ERR_PTR(-EIO);
+	}
+	lock_page(page);
+	return page;
+}
+
+/*
+ * Compare extents of two files to see if they are the same.
+ */
+static int
+xfs_compare_extents(
+	struct inode	*src,
+	xfs_off_t	srcoff,
+	struct inode	*dest,
+	xfs_off_t	destoff,
+	xfs_off_t	len,
+	bool		*is_same)
+{
+	xfs_off_t	src_poff;
+	xfs_off_t	dest_poff;
+	void		*src_addr;
+	void		*dest_addr;
+	struct page	*src_page;
+	struct page	*dest_page;
+	xfs_off_t	cmp_len;
+	bool		same;
+	int		error;
+
+	error = -EINVAL;
+	same = true;
+	while (len) {
+		src_poff = srcoff & (PAGE_SIZE - 1);
+		dest_poff = destoff & (PAGE_SIZE - 1);
+		cmp_len = min(PAGE_SIZE - src_poff,
+			      PAGE_SIZE - dest_poff);
+		cmp_len = min(cmp_len, len);
+		ASSERT(cmp_len > 0);
+
+		trace_xfs_reflink_compare_extents(XFS_I(src), srcoff, cmp_len,
+				XFS_I(dest), destoff);
+
+		src_page = xfs_get_page(src, srcoff);
+		if (IS_ERR(src_page)) {
+			error = PTR_ERR(src_page);
+			goto out_error;
+		}
+		dest_page = xfs_get_page(dest, destoff);
+		if (IS_ERR(dest_page)) {
+			error = PTR_ERR(dest_page);
+			unlock_page(src_page);
+			put_page(src_page);
+			goto out_error;
+		}
+		src_addr = kmap_atomic(src_page);
+		dest_addr = kmap_atomic(dest_page);
+
+		flush_dcache_page(src_page);
+		flush_dcache_page(dest_page);
+
+		if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
+			same = false;
+
+		kunmap_atomic(dest_addr);
+		kunmap_atomic(src_addr);
+		unlock_page(dest_page);
+		unlock_page(src_page);
+		put_page(dest_page);
+		put_page(src_page);
+
+		if (!same)
+			break;
+
+		srcoff += cmp_len;
+		destoff += cmp_len;
+		len -= cmp_len;
+	}
+
+	*is_same = same;
+	return 0;
+
+out_error:
+	trace_xfs_reflink_compare_extents_error(XFS_I(dest), error, _RET_IP_);
+	return error;
+}
+
+/*
+ * Link a range of blocks from one file to another.
+ */
+int
+xfs_reflink_remap_range(
+	struct xfs_inode	*src,
+	xfs_off_t		srcoff,
+	struct xfs_inode	*dest,
+	xfs_off_t		destoff,
+	xfs_off_t		len,
+	unsigned int		flags)
+{
+	struct xfs_mount	*mp = src->i_mount;
+	xfs_fileoff_t		sfsbno, dfsbno;
+	xfs_filblks_t		fsblen;
+	int			error;
+	xfs_extlen_t		cowextsize;
+	bool			is_same;
+
+	if (!xfs_sb_version_hasreflink(&mp->m_sb))
+		return -EOPNOTSUPP;
+
+	if (XFS_FORCED_SHUTDOWN(mp))
+		return -EIO;
+
+	/* Don't reflink realtime inodes */
+	if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
+		return -EINVAL;
+
+	if (flags & ~XFS_REFLINK_ALL)
+		return -EINVAL;
+
+	trace_xfs_reflink_remap_range(src, srcoff, len, dest, destoff);
+
+	/* Lock both files against IO */
+	if (src->i_ino == dest->i_ino) {
+		xfs_ilock(src, XFS_IOLOCK_EXCL);
+		xfs_ilock(src, XFS_MMAPLOCK_EXCL);
+	} else {
+		xfs_lock_two_inodes(src, dest, XFS_IOLOCK_EXCL);
+		xfs_lock_two_inodes(src, dest, XFS_MMAPLOCK_EXCL);
+	}
+
+	/*
+	 * Check that the extents are the same.
+	 */
+	if (flags & XFS_REFLINK_DEDUPE) {
+		is_same = false;
+		error = xfs_compare_extents(VFS_I(src), srcoff, VFS_I(dest),
+				destoff, len, &is_same);
+		if (error)
+			goto out_error;
+		if (!is_same) {
+			error = -EBADE;
+			goto out_error;
+		}
+	}
+
+	error = xfs_reflink_set_inode_flag(src, dest);
+	if (error)
+		goto out_error;
+
+	/*
+	 * Invalidate the page cache so that we can clear any CoW mappings
+	 * in the destination file.
+	 */
+	truncate_inode_pages_range(&VFS_I(dest)->i_data, destoff,
+				   PAGE_ALIGN(destoff + len) - 1);
+
+	dfsbno = XFS_B_TO_FSBT(mp, destoff);
+	sfsbno = XFS_B_TO_FSBT(mp, srcoff);
+	fsblen = XFS_B_TO_FSB(mp, len);
+	error = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen,
+			destoff + len);
+	if (error)
+		goto out_error;
+
+	/*
+	 * Carry the cowextsize hint from src to dest if we're sharing the
+	 * entire source file to the entire destination file, the source file
+	 * has a cowextsize hint, and the destination file does not.
+	 */
+	cowextsize = 0;
+	if (srcoff == 0 && len == i_size_read(VFS_I(src)) &&
+	    (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
+	    destoff == 0 && len >= i_size_read(VFS_I(dest)) &&
+	    !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
+		cowextsize = src->i_d.di_cowextsize;
+
+	error = xfs_reflink_update_dest(dest, destoff + len, cowextsize);
+	if (error)
+		goto out_error;
+
+out_error:
+	xfs_iunlock(src, XFS_MMAPLOCK_EXCL);
+	xfs_iunlock(src, XFS_IOLOCK_EXCL);
+	if (src->i_ino != dest->i_ino) {
+		xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
+		xfs_iunlock(dest, XFS_IOLOCK_EXCL);
+	}
+	if (error)
+		trace_xfs_reflink_remap_range_error(dest, error, _RET_IP_);
+	return error;
+}
+
+/*
+ * 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 (map[0].br_startblock == HOLESTARTBLOCK ||
+		    map[0].br_startblock == DELAYSTARTBLOCK ||
+		    ISUNWRITTEN(&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, 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;
+}
+
+/* Clear the inode reflink flag if there are no shared extents. */
+int
+xfs_reflink_clear_inode_flag(
+	struct xfs_inode	*ip,
+	struct xfs_trans	**tpp)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	xfs_fileoff_t		fbno;
+	xfs_filblks_t		end;
+	xfs_agnumber_t		agno;
+	xfs_agblock_t		agbno;
+	xfs_extlen_t		aglen;
+	xfs_agblock_t		rbno;
+	xfs_extlen_t		rlen;
+	struct xfs_bmbt_irec	map;
+	int			nmaps;
+	int			error = 0;
+
+	ASSERT(xfs_is_reflink_inode(ip));
+
+	fbno = 0;
+	end = XFS_B_TO_FSB(mp, i_size_read(VFS_I(ip)));
+	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)
+			return error;
+		if (nmaps == 0)
+			break;
+		if (map.br_startblock == HOLESTARTBLOCK ||
+		    map.br_startblock == DELAYSTARTBLOCK ||
+		    ISUNWRITTEN(&map))
+			goto next;
+
+		agno = XFS_FSB_TO_AGNO(mp, map.br_startblock);
+		agbno = XFS_FSB_TO_AGBNO(mp, map.br_startblock);
+		aglen = map.br_blockcount;
+
+		error = xfs_reflink_find_shared(mp, agno, agbno, aglen,
+				&rbno, &rlen, false);
+		if (error)
+			return error;
+		/* Is there still a shared block here? */
+		if (rbno != NULLAGBLOCK)
+			return 0;
+next:
+		fbno = map.br_startoff + map.br_blockcount;
+	}
+
+	/*
+	 * 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);
+	if (error)
+		return error;
+
+	/* Clear the inode flag. */
+	trace_xfs_reflink_unset_inode_flag(ip);
+	ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
+	xfs_inode_clear_cowblocks_tag(ip);
+	xfs_trans_ijoin(*tpp, ip, 0);
+	xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
+
+	return error;
+}
+
+/*
+ * Clear the inode reflink flag if there are no shared extents and the size
+ * hasn't changed.
+ */
+STATIC int
+xfs_reflink_try_clear_inode_flag(
+	struct xfs_inode	*ip)
+{
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_trans	*tp;
+	int			error = 0;
+
+	/* Start a rolling transaction to remove the mappings */
+	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
+	if (error)
+		return error;
+
+	xfs_ilock(ip, XFS_ILOCK_EXCL);
+	xfs_trans_ijoin(tp, ip, 0);
+
+	error = xfs_reflink_clear_inode_flag(ip, &tp);
+	if (error)
+		goto cancel;
+
+	error = xfs_trans_commit(tp);
+	if (error)
+		goto out;
+
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	return 0;
+cancel:
+	xfs_trans_cancel(tp);
+out:
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+	return error;
+}
+
+/*
+ * Pre-COW all shared blocks within a given byte range of a file and turn off
+ * the reflink flag if we unshare all of the file's blocks.
+ */
+int
+xfs_reflink_unshare(
+	struct xfs_inode	*ip,
+	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;
+	int			error;
+
+	if (!xfs_is_reflink_inode(ip))
+		return 0;
+
+	trace_xfs_reflink_unshare(ip, offset, len);
+
+	inode_dio_wait(VFS_I(ip));
+
+	/* 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 (error)
+		goto out_unlock;
+	xfs_iunlock(ip, XFS_ILOCK_EXCL);
+
+	/* Wait for the IO to finish */
+	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
+	if (error)
+		goto out;
+
+	/* Turn off the reflink flag if possible. */
+	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;
+}
+
+/*
+ * Does this inode have any real CoW reservations?
+ */
+bool
+xfs_reflink_has_real_cow_blocks(
+	struct xfs_inode		*ip)
+{
+	struct xfs_bmbt_irec		irec;
+	struct xfs_ifork		*ifp;
+	struct xfs_bmbt_rec_host	*gotp;
+	xfs_extnum_t			idx;
+
+	if (!xfs_is_reflink_inode(ip))
+		return false;
+
+	/* Go find the old extent in the CoW fork. */
+	ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
+	gotp = xfs_iext_bno_to_ext(ifp, 0, &idx);
+	while (gotp) {
+		xfs_bmbt_get_all(gotp, &irec);
+
+		if (!isnullstartblock(irec.br_startblock))
+			return true;
+
+		/* Roll on... */
+		idx++;
+		if (idx >= ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
+			break;
+		gotp = xfs_iext_get_ext(ifp, idx);
+	}
+
+	return false;
+}

fs/xfs/xfs_reflink.h

@@ -0,0 +1,58 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_REFLINK_H
+#define __XFS_REFLINK_H 1
+
+extern int xfs_reflink_find_shared(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agblock_t agbno, xfs_extlen_t aglen, xfs_agblock_t *fbno,
+		xfs_extlen_t *flen, bool find_maximal);
+extern int xfs_reflink_trim_around_shared(struct xfs_inode *ip,
+		struct xfs_bmbt_irec *irec, bool *shared, bool *trimmed);
+
+extern int xfs_reflink_reserve_cow_range(struct xfs_inode *ip,
+		xfs_off_t offset, xfs_off_t count);
+extern int xfs_reflink_allocate_cow_range(struct xfs_inode *ip,
+		xfs_off_t offset, xfs_off_t count);
+extern bool xfs_reflink_find_cow_mapping(struct xfs_inode *ip, xfs_off_t offset,
+		struct xfs_bmbt_irec *imap, bool *need_alloc);
+extern int xfs_reflink_trim_irec_to_next_cow(struct xfs_inode *ip,
+		xfs_fileoff_t offset_fsb, struct xfs_bmbt_irec *imap);
+
+extern int xfs_reflink_cancel_cow_blocks(struct xfs_inode *ip,
+		struct xfs_trans **tpp, xfs_fileoff_t offset_fsb,
+		xfs_fileoff_t end_fsb);
+extern int xfs_reflink_cancel_cow_range(struct xfs_inode *ip, xfs_off_t offset,
+		xfs_off_t count);
+extern int xfs_reflink_end_cow(struct xfs_inode *ip, xfs_off_t offset,
+		xfs_off_t count);
+extern int xfs_reflink_recover_cow(struct xfs_mount *mp);
+#define XFS_REFLINK_DEDUPE	1	/* only reflink if contents match */
+#define XFS_REFLINK_ALL		(XFS_REFLINK_DEDUPE)
+extern int xfs_reflink_remap_range(struct xfs_inode *src, xfs_off_t srcoff,
+		struct xfs_inode *dest, xfs_off_t destoff, xfs_off_t len,
+		unsigned int flags);
+extern int xfs_reflink_clear_inode_flag(struct xfs_inode *ip,
+		struct xfs_trans **tpp);
+extern int xfs_reflink_unshare(struct xfs_inode *ip, xfs_off_t offset,
+		xfs_off_t len);
+
+extern bool xfs_reflink_has_real_cow_blocks(struct xfs_inode *ip);
+
+#endif /* __XFS_REFLINK_H */

fs/xfs/xfs_rmap_item.c

@@ -441,8 +441,11 @@ xfs_rui_recover(
 				   XFS_FSB_TO_DADDR(mp, rmap->me_startblock));
 		switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
 		case XFS_RMAP_EXTENT_MAP:
+		case XFS_RMAP_EXTENT_MAP_SHARED:
 		case XFS_RMAP_EXTENT_UNMAP:
+		case XFS_RMAP_EXTENT_UNMAP_SHARED:
 		case XFS_RMAP_EXTENT_CONVERT:
+		case XFS_RMAP_EXTENT_CONVERT_SHARED:
 		case XFS_RMAP_EXTENT_ALLOC:
 		case XFS_RMAP_EXTENT_FREE:
 			op_ok = true;
@@ -481,12 +484,21 @@ xfs_rui_recover(
 		case XFS_RMAP_EXTENT_MAP:
 			type = XFS_RMAP_MAP;
 			break;
+		case XFS_RMAP_EXTENT_MAP_SHARED:
+			type = XFS_RMAP_MAP_SHARED;
+			break;
 		case XFS_RMAP_EXTENT_UNMAP:
 			type = XFS_RMAP_UNMAP;
 			break;
+		case XFS_RMAP_EXTENT_UNMAP_SHARED:
+			type = XFS_RMAP_UNMAP_SHARED;
+			break;
 		case XFS_RMAP_EXTENT_CONVERT:
 			type = XFS_RMAP_CONVERT;
 			break;
+		case XFS_RMAP_EXTENT_CONVERT_SHARED:
+			type = XFS_RMAP_CONVERT_SHARED;
+			break;
 		case XFS_RMAP_EXTENT_ALLOC:
 			type = XFS_RMAP_ALLOC;
 			break;

fs/xfs/xfs_stats.c

@@ -62,6 +62,7 @@ int xfs_stats_format(struct xfsstats __percpu *stats, char *buf)
 		{ "ibt2",		XFSSTAT_END_IBT_V2		},
 		{ "fibt2",		XFSSTAT_END_FIBT_V2		},
 		{ "rmapbt",		XFSSTAT_END_RMAP_V2		},
+		{ "refcntbt",		XFSSTAT_END_REFCOUNT		},
 		/* we print both series of quota information together */
 		{ "qm",			XFSSTAT_END_QM			},
 	};

fs/xfs/xfs_stats.h

@@ -213,7 +213,23 @@ struct xfsstats {
 	__uint32_t		xs_rmap_2_alloc;
 	__uint32_t		xs_rmap_2_free;
 	__uint32_t		xs_rmap_2_moves;
-#define XFSSTAT_END_XQMSTAT		(XFSSTAT_END_RMAP_V2+6)
+#define XFSSTAT_END_REFCOUNT		(XFSSTAT_END_RMAP_V2 + 15)
+	__uint32_t		xs_refcbt_2_lookup;
+	__uint32_t		xs_refcbt_2_compare;
+	__uint32_t		xs_refcbt_2_insrec;
+	__uint32_t		xs_refcbt_2_delrec;
+	__uint32_t		xs_refcbt_2_newroot;
+	__uint32_t		xs_refcbt_2_killroot;
+	__uint32_t		xs_refcbt_2_increment;
+	__uint32_t		xs_refcbt_2_decrement;
+	__uint32_t		xs_refcbt_2_lshift;
+	__uint32_t		xs_refcbt_2_rshift;
+	__uint32_t		xs_refcbt_2_split;
+	__uint32_t		xs_refcbt_2_join;
+	__uint32_t		xs_refcbt_2_alloc;
+	__uint32_t		xs_refcbt_2_free;
+	__uint32_t		xs_refcbt_2_moves;
+#define XFSSTAT_END_XQMSTAT		(XFSSTAT_END_REFCOUNT + 6)
 	__uint32_t		xs_qm_dqreclaims;
 	__uint32_t		xs_qm_dqreclaim_misses;
 	__uint32_t		xs_qm_dquot_dups;

fs/xfs/xfs_super.c

@@ -47,6 +47,9 @@
 #include "xfs_sysfs.h"
 #include "xfs_ondisk.h"
 #include "xfs_rmap_item.h"
+#include "xfs_refcount_item.h"
+#include "xfs_bmap_item.h"
+#include "xfs_reflink.h"
 
 #include <linux/namei.h>
 #include <linux/init.h>
@@ -936,6 +939,7 @@ xfs_fs_destroy_inode(
 	struct inode		*inode)
 {
 	struct xfs_inode	*ip = XFS_I(inode);
+	int			error;
 
 	trace_xfs_destroy_inode(ip);
 
@@ -943,6 +947,14 @@ xfs_fs_destroy_inode(
 	XFS_STATS_INC(ip->i_mount, vn_rele);
 	XFS_STATS_INC(ip->i_mount, vn_remove);
 
+	if (xfs_is_reflink_inode(ip)) {
+		error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF);
+		if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
+			xfs_warn(ip->i_mount,
+"Error %d while evicting CoW blocks for inode %llu.",
+					error, ip->i_ino);
+	}
+
 	xfs_inactive(ip);
 
 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
@@ -1006,6 +1018,16 @@ xfs_fs_drop_inode(
 {
 	struct xfs_inode	*ip = XFS_I(inode);
 
+	/*
+	 * If this unlinked inode is in the middle of recovery, don't
+	 * drop the inode just yet; log recovery will take care of
+	 * that.  See the comment for this inode flag.
+	 */
+	if (ip->i_flags & XFS_IRECOVERY) {
+		ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
+		return 0;
+	}
+
 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
 }
 
@@ -1296,10 +1318,31 @@ xfs_fs_remount(
 		xfs_restore_resvblks(mp);
 		xfs_log_work_queue(mp);
 		xfs_queue_eofblocks(mp);
+
+		/* Recover any CoW blocks that never got remapped. */
+		error = xfs_reflink_recover_cow(mp);
+		if (error) {
+			xfs_err(mp,
+	"Error %d recovering leftover CoW allocations.", error);
+			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+			return error;
+		}
+
+		/* Create the per-AG metadata reservation pool .*/
+		error = xfs_fs_reserve_ag_blocks(mp);
+		if (error && error != -ENOSPC)
+			return error;
 	}
 
 	/* rw -> ro */
 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
+		/* Free the per-AG metadata reservation pool. */
+		error = xfs_fs_unreserve_ag_blocks(mp);
+		if (error) {
+			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+			return error;
+		}
+
 		/*
 		 * Before we sync the metadata, we need to free up the reserve
 		 * block pool so that the used block count in the superblock on
@@ -1490,6 +1533,7 @@ xfs_fs_fill_super(
 	atomic_set(&mp->m_active_trans, 0);
 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
+	INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
 	mp->m_kobj.kobject.kset = xfs_kset;
 
 	mp->m_super = sb;
@@ -1572,6 +1616,9 @@ xfs_fs_fill_super(
 			"DAX unsupported by block device. Turning off DAX.");
 			mp->m_flags &= ~XFS_MOUNT_DAX;
 		}
+		if (xfs_sb_version_hasreflink(&mp->m_sb))
+			xfs_alert(mp,
+		"DAX and reflink have not been tested together!");
 	}
 
 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
@@ -1585,6 +1632,10 @@ xfs_fs_fill_super(
 	"EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
 	}
 
+	if (xfs_sb_version_hasreflink(&mp->m_sb))
+		xfs_alert(mp,
+	"EXPERIMENTAL reflink feature enabled. Use at your own risk!");
+
 	error = xfs_mountfs(mp);
 	if (error)
 		goto out_filestream_unmount;
@@ -1788,8 +1839,38 @@ xfs_init_zones(void)
 	if (!xfs_rui_zone)
 		goto out_destroy_rud_zone;
 
+	xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
+			"xfs_cud_item");
+	if (!xfs_cud_zone)
+		goto out_destroy_rui_zone;
+
+	xfs_cui_zone = kmem_zone_init(
+			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
+			"xfs_cui_item");
+	if (!xfs_cui_zone)
+		goto out_destroy_cud_zone;
+
+	xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
+			"xfs_bud_item");
+	if (!xfs_bud_zone)
+		goto out_destroy_cui_zone;
+
+	xfs_bui_zone = kmem_zone_init(
+			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
+			"xfs_bui_item");
+	if (!xfs_bui_zone)
+		goto out_destroy_bud_zone;
+
 	return 0;
 
+ out_destroy_bud_zone:
+	kmem_zone_destroy(xfs_bud_zone);
+ out_destroy_cui_zone:
+	kmem_zone_destroy(xfs_cui_zone);
+ out_destroy_cud_zone:
+	kmem_zone_destroy(xfs_cud_zone);
+ out_destroy_rui_zone:
+	kmem_zone_destroy(xfs_rui_zone);
  out_destroy_rud_zone:
 	kmem_zone_destroy(xfs_rud_zone);
  out_destroy_icreate_zone:
@@ -1832,6 +1913,10 @@ xfs_destroy_zones(void)
 	 * destroy caches.
 	 */
 	rcu_barrier();
+	kmem_zone_destroy(xfs_bui_zone);
+	kmem_zone_destroy(xfs_bud_zone);
+	kmem_zone_destroy(xfs_cui_zone);
+	kmem_zone_destroy(xfs_cud_zone);
 	kmem_zone_destroy(xfs_rui_zone);
 	kmem_zone_destroy(xfs_rud_zone);
 	kmem_zone_destroy(xfs_icreate_zone);
@@ -1885,6 +1970,8 @@ init_xfs_fs(void)
 
 	xfs_extent_free_init_defer_op();
 	xfs_rmap_update_init_defer_op();
+	xfs_refcount_update_init_defer_op();
+	xfs_bmap_update_init_defer_op();
 
 	xfs_dir_startup();
 

fs/xfs/xfs_sysctl.c

@@ -184,6 +184,15 @@ static struct ctl_table xfs_table[] = {
 		.extra1		= &xfs_params.eofb_timer.min,
 		.extra2		= &xfs_params.eofb_timer.max,
 	},
+	{
+		.procname	= "speculative_cow_prealloc_lifetime",
+		.data		= &xfs_params.cowb_timer.val,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &xfs_params.cowb_timer.min,
+		.extra2		= &xfs_params.cowb_timer.max,
+	},
 	/* please keep this the last entry */
 #ifdef CONFIG_PROC_FS
 	{

fs/xfs/xfs_sysctl.h

@@ -48,6 +48,7 @@ typedef struct xfs_param {
 	xfs_sysctl_val_t inherit_nodfrg;/* Inherit the "nodefrag" inode flag. */
 	xfs_sysctl_val_t fstrm_timer;	/* Filestream dir-AG assoc'n timeout. */
 	xfs_sysctl_val_t eofb_timer;	/* Interval between eofb scan wakeups */
+	xfs_sysctl_val_t cowb_timer;	/* Interval between cowb scan wakeups */
 } xfs_param_t;
 
 /*

fs/xfs/xfs_trace.h

@@ -39,6 +39,7 @@ struct xfs_buf_log_format;
 struct xfs_inode_log_format;
 struct xfs_bmbt_irec;
 struct xfs_btree_cur;
+struct xfs_refcount_irec;
 
 DECLARE_EVENT_CLASS(xfs_attr_list_class,
 	TP_PROTO(struct xfs_attr_list_context *ctx),
@@ -135,6 +136,8 @@ DEFINE_PERAG_REF_EVENT(xfs_perag_set_reclaim);
 DEFINE_PERAG_REF_EVENT(xfs_perag_clear_reclaim);
 DEFINE_PERAG_REF_EVENT(xfs_perag_set_eofblocks);
 DEFINE_PERAG_REF_EVENT(xfs_perag_clear_eofblocks);
+DEFINE_PERAG_REF_EVENT(xfs_perag_set_cowblocks);
+DEFINE_PERAG_REF_EVENT(xfs_perag_clear_cowblocks);
 
 DECLARE_EVENT_CLASS(xfs_ag_class,
 	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno),
@@ -268,10 +271,10 @@ DECLARE_EVENT_CLASS(xfs_bmap_class,
 		__field(unsigned long, caller_ip)
 	),
 	TP_fast_assign(
-		struct xfs_ifork	*ifp = (state & BMAP_ATTRFORK) ?
-						ip->i_afp : &ip->i_df;
+		struct xfs_ifork	*ifp;
 		struct xfs_bmbt_irec	r;
 
+		ifp = xfs_iext_state_to_fork(ip, state);
 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &r);
 		__entry->dev = VFS_I(ip)->i_sb->s_dev;
 		__entry->ino = ip->i_ino;
@@ -686,6 +689,9 @@ DEFINE_INODE_EVENT(xfs_dquot_dqdetach);
 DEFINE_INODE_EVENT(xfs_inode_set_eofblocks_tag);
 DEFINE_INODE_EVENT(xfs_inode_clear_eofblocks_tag);
 DEFINE_INODE_EVENT(xfs_inode_free_eofblocks_invalid);
+DEFINE_INODE_EVENT(xfs_inode_set_cowblocks_tag);
+DEFINE_INODE_EVENT(xfs_inode_clear_cowblocks_tag);
+DEFINE_INODE_EVENT(xfs_inode_free_cowblocks_invalid);
 
 DEFINE_INODE_EVENT(xfs_filemap_fault);
 DEFINE_INODE_EVENT(xfs_filemap_pmd_fault);
@@ -2581,10 +2587,20 @@ DEFINE_RMAPBT_EVENT(xfs_rmap_delete);
 DEFINE_AG_ERROR_EVENT(xfs_rmap_insert_error);
 DEFINE_AG_ERROR_EVENT(xfs_rmap_delete_error);
 DEFINE_AG_ERROR_EVENT(xfs_rmap_update_error);
+
+DEFINE_RMAPBT_EVENT(xfs_rmap_find_left_neighbor_candidate);
+DEFINE_RMAPBT_EVENT(xfs_rmap_find_left_neighbor_query);
+DEFINE_RMAPBT_EVENT(xfs_rmap_lookup_le_range_candidate);
+DEFINE_RMAPBT_EVENT(xfs_rmap_lookup_le_range);
 DEFINE_RMAPBT_EVENT(xfs_rmap_lookup_le_range_result);
 DEFINE_RMAPBT_EVENT(xfs_rmap_find_right_neighbor_result);
 DEFINE_RMAPBT_EVENT(xfs_rmap_find_left_neighbor_result);
 
+/* deferred bmbt updates */
+#define DEFINE_BMAP_DEFERRED_EVENT	DEFINE_RMAP_DEFERRED_EVENT
+DEFINE_BMAP_DEFERRED_EVENT(xfs_bmap_defer);
+DEFINE_BMAP_DEFERRED_EVENT(xfs_bmap_deferred);
+
 /* per-AG reservation */
 DECLARE_EVENT_CLASS(xfs_ag_resv_class,
 	TP_PROTO(struct xfs_perag *pag, enum xfs_ag_resv_type resv,
@@ -2639,6 +2655,728 @@ DEFINE_AG_RESV_EVENT(xfs_ag_resv_needed);
 DEFINE_AG_ERROR_EVENT(xfs_ag_resv_free_error);
 DEFINE_AG_ERROR_EVENT(xfs_ag_resv_init_error);
 
+/* refcount tracepoint classes */
+
+/* reuse the discard trace class for agbno/aglen-based traces */
+#define DEFINE_AG_EXTENT_EVENT(name) DEFINE_DISCARD_EVENT(name)
+
+/* ag btree lookup tracepoint class */
+#define XFS_AG_BTREE_CMP_FORMAT_STR \
+	{ XFS_LOOKUP_EQ,	"eq" }, \
+	{ XFS_LOOKUP_LE,	"le" }, \
+	{ XFS_LOOKUP_GE,	"ge" }
+DECLARE_EVENT_CLASS(xfs_ag_btree_lookup_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 xfs_agblock_t agbno, xfs_lookup_t dir),
+	TP_ARGS(mp, agno, agbno, dir),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, agbno)
+		__field(xfs_lookup_t, dir)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->agbno = agbno;
+		__entry->dir = dir;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u cmp %s(%d)\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->agbno,
+		  __print_symbolic(__entry->dir, XFS_AG_BTREE_CMP_FORMAT_STR),
+		  __entry->dir)
+)
+
+#define DEFINE_AG_BTREE_LOOKUP_EVENT(name) \
+DEFINE_EVENT(xfs_ag_btree_lookup_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 xfs_agblock_t agbno, xfs_lookup_t dir), \
+	TP_ARGS(mp, agno, agbno, dir))
+
+/* single-rcext tracepoint class */
+DECLARE_EVENT_CLASS(xfs_refcount_extent_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 struct xfs_refcount_irec *irec),
+	TP_ARGS(mp, agno, irec),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, startblock)
+		__field(xfs_extlen_t, blockcount)
+		__field(xfs_nlink_t, refcount)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->startblock = irec->rc_startblock;
+		__entry->blockcount = irec->rc_blockcount;
+		__entry->refcount = irec->rc_refcount;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u len %u refcount %u\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->startblock,
+		  __entry->blockcount,
+		  __entry->refcount)
+)
+
+#define DEFINE_REFCOUNT_EXTENT_EVENT(name) \
+DEFINE_EVENT(xfs_refcount_extent_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 struct xfs_refcount_irec *irec), \
+	TP_ARGS(mp, agno, irec))
+
+/* single-rcext and an agbno tracepoint class */
+DECLARE_EVENT_CLASS(xfs_refcount_extent_at_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 struct xfs_refcount_irec *irec, xfs_agblock_t agbno),
+	TP_ARGS(mp, agno, irec, agbno),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, startblock)
+		__field(xfs_extlen_t, blockcount)
+		__field(xfs_nlink_t, refcount)
+		__field(xfs_agblock_t, agbno)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->startblock = irec->rc_startblock;
+		__entry->blockcount = irec->rc_blockcount;
+		__entry->refcount = irec->rc_refcount;
+		__entry->agbno = agbno;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u len %u refcount %u @ agbno %u\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->startblock,
+		  __entry->blockcount,
+		  __entry->refcount,
+		  __entry->agbno)
+)
+
+#define DEFINE_REFCOUNT_EXTENT_AT_EVENT(name) \
+DEFINE_EVENT(xfs_refcount_extent_at_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 struct xfs_refcount_irec *irec, xfs_agblock_t agbno), \
+	TP_ARGS(mp, agno, irec, agbno))
+
+/* double-rcext tracepoint class */
+DECLARE_EVENT_CLASS(xfs_refcount_double_extent_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2),
+	TP_ARGS(mp, agno, i1, i2),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, i1_startblock)
+		__field(xfs_extlen_t, i1_blockcount)
+		__field(xfs_nlink_t, i1_refcount)
+		__field(xfs_agblock_t, i2_startblock)
+		__field(xfs_extlen_t, i2_blockcount)
+		__field(xfs_nlink_t, i2_refcount)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->i1_startblock = i1->rc_startblock;
+		__entry->i1_blockcount = i1->rc_blockcount;
+		__entry->i1_refcount = i1->rc_refcount;
+		__entry->i2_startblock = i2->rc_startblock;
+		__entry->i2_blockcount = i2->rc_blockcount;
+		__entry->i2_refcount = i2->rc_refcount;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u len %u refcount %u -- "
+		  "agbno %u len %u refcount %u\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->i1_startblock,
+		  __entry->i1_blockcount,
+		  __entry->i1_refcount,
+		  __entry->i2_startblock,
+		  __entry->i2_blockcount,
+		  __entry->i2_refcount)
+)
+
+#define DEFINE_REFCOUNT_DOUBLE_EXTENT_EVENT(name) \
+DEFINE_EVENT(xfs_refcount_double_extent_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2), \
+	TP_ARGS(mp, agno, i1, i2))
+
+/* double-rcext and an agbno tracepoint class */
+DECLARE_EVENT_CLASS(xfs_refcount_double_extent_at_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2,
+		 xfs_agblock_t agbno),
+	TP_ARGS(mp, agno, i1, i2, agbno),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, i1_startblock)
+		__field(xfs_extlen_t, i1_blockcount)
+		__field(xfs_nlink_t, i1_refcount)
+		__field(xfs_agblock_t, i2_startblock)
+		__field(xfs_extlen_t, i2_blockcount)
+		__field(xfs_nlink_t, i2_refcount)
+		__field(xfs_agblock_t, agbno)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->i1_startblock = i1->rc_startblock;
+		__entry->i1_blockcount = i1->rc_blockcount;
+		__entry->i1_refcount = i1->rc_refcount;
+		__entry->i2_startblock = i2->rc_startblock;
+		__entry->i2_blockcount = i2->rc_blockcount;
+		__entry->i2_refcount = i2->rc_refcount;
+		__entry->agbno = agbno;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u len %u refcount %u -- "
+		  "agbno %u len %u refcount %u @ agbno %u\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->i1_startblock,
+		  __entry->i1_blockcount,
+		  __entry->i1_refcount,
+		  __entry->i2_startblock,
+		  __entry->i2_blockcount,
+		  __entry->i2_refcount,
+		  __entry->agbno)
+)
+
+#define DEFINE_REFCOUNT_DOUBLE_EXTENT_AT_EVENT(name) \
+DEFINE_EVENT(xfs_refcount_double_extent_at_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2, \
+		 xfs_agblock_t agbno), \
+	TP_ARGS(mp, agno, i1, i2, agbno))
+
+/* triple-rcext tracepoint class */
+DECLARE_EVENT_CLASS(xfs_refcount_triple_extent_class,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2,
+		 struct xfs_refcount_irec *i3),
+	TP_ARGS(mp, agno, i1, i2, i3),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_agblock_t, i1_startblock)
+		__field(xfs_extlen_t, i1_blockcount)
+		__field(xfs_nlink_t, i1_refcount)
+		__field(xfs_agblock_t, i2_startblock)
+		__field(xfs_extlen_t, i2_blockcount)
+		__field(xfs_nlink_t, i2_refcount)
+		__field(xfs_agblock_t, i3_startblock)
+		__field(xfs_extlen_t, i3_blockcount)
+		__field(xfs_nlink_t, i3_refcount)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->i1_startblock = i1->rc_startblock;
+		__entry->i1_blockcount = i1->rc_blockcount;
+		__entry->i1_refcount = i1->rc_refcount;
+		__entry->i2_startblock = i2->rc_startblock;
+		__entry->i2_blockcount = i2->rc_blockcount;
+		__entry->i2_refcount = i2->rc_refcount;
+		__entry->i3_startblock = i3->rc_startblock;
+		__entry->i3_blockcount = i3->rc_blockcount;
+		__entry->i3_refcount = i3->rc_refcount;
+	),
+	TP_printk("dev %d:%d agno %u agbno %u len %u refcount %u -- "
+		  "agbno %u len %u refcount %u -- "
+		  "agbno %u len %u refcount %u\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->agno,
+		  __entry->i1_startblock,
+		  __entry->i1_blockcount,
+		  __entry->i1_refcount,
+		  __entry->i2_startblock,
+		  __entry->i2_blockcount,
+		  __entry->i2_refcount,
+		  __entry->i3_startblock,
+		  __entry->i3_blockcount,
+		  __entry->i3_refcount)
+);
+
+#define DEFINE_REFCOUNT_TRIPLE_EXTENT_EVENT(name) \
+DEFINE_EVENT(xfs_refcount_triple_extent_class, name, \
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
+		 struct xfs_refcount_irec *i1, struct xfs_refcount_irec *i2, \
+		 struct xfs_refcount_irec *i3), \
+	TP_ARGS(mp, agno, i1, i2, i3))
+
+/* refcount btree tracepoints */
+DEFINE_BUSY_EVENT(xfs_refcountbt_alloc_block);
+DEFINE_BUSY_EVENT(xfs_refcountbt_free_block);
+DEFINE_AG_BTREE_LOOKUP_EVENT(xfs_refcount_lookup);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_get);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_update);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_insert);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_delete);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_insert_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_delete_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_update_error);
+
+/* refcount adjustment tracepoints */
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_increase);
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_decrease);
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_cow_increase);
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_cow_decrease);
+DEFINE_REFCOUNT_TRIPLE_EXTENT_EVENT(xfs_refcount_merge_center_extents);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_modify_extent);
+DEFINE_REFCOUNT_EXTENT_EVENT(xfs_refcount_recover_extent);
+DEFINE_REFCOUNT_EXTENT_AT_EVENT(xfs_refcount_split_extent);
+DEFINE_REFCOUNT_DOUBLE_EXTENT_EVENT(xfs_refcount_merge_left_extent);
+DEFINE_REFCOUNT_DOUBLE_EXTENT_EVENT(xfs_refcount_merge_right_extent);
+DEFINE_REFCOUNT_DOUBLE_EXTENT_AT_EVENT(xfs_refcount_find_left_extent);
+DEFINE_REFCOUNT_DOUBLE_EXTENT_AT_EVENT(xfs_refcount_find_right_extent);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_adjust_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_adjust_cow_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_merge_center_extents_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_modify_extent_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_split_extent_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_merge_left_extent_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_merge_right_extent_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_find_left_extent_error);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_find_right_extent_error);
+
+/* reflink helpers */
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_find_shared);
+DEFINE_AG_EXTENT_EVENT(xfs_refcount_find_shared_result);
+DEFINE_AG_ERROR_EVENT(xfs_refcount_find_shared_error);
+#define DEFINE_REFCOUNT_DEFERRED_EVENT DEFINE_PHYS_EXTENT_DEFERRED_EVENT
+DEFINE_REFCOUNT_DEFERRED_EVENT(xfs_refcount_defer);
+DEFINE_REFCOUNT_DEFERRED_EVENT(xfs_refcount_deferred);
+
+TRACE_EVENT(xfs_refcount_finish_one_leftover,
+	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
+		 int type, xfs_agblock_t agbno, xfs_extlen_t len,
+		 xfs_agblock_t new_agbno, xfs_extlen_t new_len),
+	TP_ARGS(mp, agno, type, agbno, len, new_agbno, new_len),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_agnumber_t, agno)
+		__field(int, type)
+		__field(xfs_agblock_t, agbno)
+		__field(xfs_extlen_t, len)
+		__field(xfs_agblock_t, new_agbno)
+		__field(xfs_extlen_t, new_len)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->agno = agno;
+		__entry->type = type;
+		__entry->agbno = agbno;
+		__entry->len = len;
+		__entry->new_agbno = new_agbno;
+		__entry->new_len = new_len;
+	),
+	TP_printk("dev %d:%d type %d agno %u agbno %u len %u new_agbno %u new_len %u",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->type,
+		  __entry->agno,
+		  __entry->agbno,
+		  __entry->len,
+		  __entry->new_agbno,
+		  __entry->new_len)
+);
+
+/* simple inode-based error/%ip tracepoint class */
+DECLARE_EVENT_CLASS(xfs_inode_error_class,
+	TP_PROTO(struct xfs_inode *ip, int error, unsigned long caller_ip),
+	TP_ARGS(ip, error, caller_ip),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(int, error)
+		__field(unsigned long, caller_ip)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->error = error;
+		__entry->caller_ip = caller_ip;
+	),
+	TP_printk("dev %d:%d ino %llx error %d caller %ps",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->error,
+		  (char *)__entry->caller_ip)
+);
+
+#define DEFINE_INODE_ERROR_EVENT(name) \
+DEFINE_EVENT(xfs_inode_error_class, name, \
+	TP_PROTO(struct xfs_inode *ip, int error, \
+		 unsigned long caller_ip), \
+	TP_ARGS(ip, error, caller_ip))
+
+/* reflink allocator */
+TRACE_EVENT(xfs_bmap_remap_alloc,
+	TP_PROTO(struct xfs_inode *ip, xfs_fsblock_t fsbno,
+		 xfs_extlen_t len),
+	TP_ARGS(ip, fsbno, len),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_fsblock_t, fsbno)
+		__field(xfs_extlen_t, len)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->fsbno = fsbno;
+		__entry->len = len;
+	),
+	TP_printk("dev %d:%d ino 0x%llx fsbno 0x%llx len %x",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->fsbno,
+		  __entry->len)
+);
+DEFINE_INODE_ERROR_EVENT(xfs_bmap_remap_alloc_error);
+
+/* reflink tracepoint classes */
+
+/* two-file io tracepoint class */
+DECLARE_EVENT_CLASS(xfs_double_io_class,
+	TP_PROTO(struct xfs_inode *src, xfs_off_t soffset, xfs_off_t len,
+		 struct xfs_inode *dest, xfs_off_t doffset),
+	TP_ARGS(src, soffset, len, dest, doffset),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, src_ino)
+		__field(loff_t, src_isize)
+		__field(loff_t, src_disize)
+		__field(loff_t, src_offset)
+		__field(size_t, len)
+		__field(xfs_ino_t, dest_ino)
+		__field(loff_t, dest_isize)
+		__field(loff_t, dest_disize)
+		__field(loff_t, dest_offset)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(src)->i_sb->s_dev;
+		__entry->src_ino = src->i_ino;
+		__entry->src_isize = VFS_I(src)->i_size;
+		__entry->src_disize = src->i_d.di_size;
+		__entry->src_offset = soffset;
+		__entry->len = len;
+		__entry->dest_ino = dest->i_ino;
+		__entry->dest_isize = VFS_I(dest)->i_size;
+		__entry->dest_disize = dest->i_d.di_size;
+		__entry->dest_offset = doffset;
+	),
+	TP_printk("dev %d:%d count %zd "
+		  "ino 0x%llx isize 0x%llx disize 0x%llx offset 0x%llx -> "
+		  "ino 0x%llx isize 0x%llx disize 0x%llx offset 0x%llx",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->len,
+		  __entry->src_ino,
+		  __entry->src_isize,
+		  __entry->src_disize,
+		  __entry->src_offset,
+		  __entry->dest_ino,
+		  __entry->dest_isize,
+		  __entry->dest_disize,
+		  __entry->dest_offset)
+)
+
+#define DEFINE_DOUBLE_IO_EVENT(name)	\
+DEFINE_EVENT(xfs_double_io_class, name,	\
+	TP_PROTO(struct xfs_inode *src, xfs_off_t soffset, xfs_off_t len, \
+		 struct xfs_inode *dest, xfs_off_t doffset), \
+	TP_ARGS(src, soffset, len, dest, doffset))
+
+/* two-file vfs io tracepoint class */
+DECLARE_EVENT_CLASS(xfs_double_vfs_io_class,
+	TP_PROTO(struct inode *src, u64 soffset, u64 len,
+		 struct inode *dest, u64 doffset),
+	TP_ARGS(src, soffset, len, dest, doffset),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(unsigned long, src_ino)
+		__field(loff_t, src_isize)
+		__field(loff_t, src_offset)
+		__field(size_t, len)
+		__field(unsigned long, dest_ino)
+		__field(loff_t, dest_isize)
+		__field(loff_t, dest_offset)
+	),
+	TP_fast_assign(
+		__entry->dev = src->i_sb->s_dev;
+		__entry->src_ino = src->i_ino;
+		__entry->src_isize = i_size_read(src);
+		__entry->src_offset = soffset;
+		__entry->len = len;
+		__entry->dest_ino = dest->i_ino;
+		__entry->dest_isize = i_size_read(dest);
+		__entry->dest_offset = doffset;
+	),
+	TP_printk("dev %d:%d count %zd "
+		  "ino 0x%lx isize 0x%llx offset 0x%llx -> "
+		  "ino 0x%lx isize 0x%llx offset 0x%llx",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->len,
+		  __entry->src_ino,
+		  __entry->src_isize,
+		  __entry->src_offset,
+		  __entry->dest_ino,
+		  __entry->dest_isize,
+		  __entry->dest_offset)
+)
+
+#define DEFINE_DOUBLE_VFS_IO_EVENT(name)	\
+DEFINE_EVENT(xfs_double_vfs_io_class, name,	\
+	TP_PROTO(struct inode *src, u64 soffset, u64 len, \
+		 struct inode *dest, u64 doffset), \
+	TP_ARGS(src, soffset, len, dest, doffset))
+
+/* CoW write tracepoint */
+DECLARE_EVENT_CLASS(xfs_copy_on_write_class,
+	TP_PROTO(struct xfs_inode *ip, xfs_fileoff_t lblk, xfs_fsblock_t pblk,
+		 xfs_extlen_t len, xfs_fsblock_t new_pblk),
+	TP_ARGS(ip, lblk, pblk, len, new_pblk),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_fileoff_t, lblk)
+		__field(xfs_fsblock_t, pblk)
+		__field(xfs_extlen_t, len)
+		__field(xfs_fsblock_t, new_pblk)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->lblk = lblk;
+		__entry->pblk = pblk;
+		__entry->len = len;
+		__entry->new_pblk = new_pblk;
+	),
+	TP_printk("dev %d:%d ino 0x%llx lblk 0x%llx pblk 0x%llx "
+		  "len 0x%x new_pblk %llu",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->lblk,
+		  __entry->pblk,
+		  __entry->len,
+		  __entry->new_pblk)
+)
+
+#define DEFINE_COW_EVENT(name)	\
+DEFINE_EVENT(xfs_copy_on_write_class, name,	\
+	TP_PROTO(struct xfs_inode *ip, xfs_fileoff_t lblk, xfs_fsblock_t pblk, \
+		 xfs_extlen_t len, xfs_fsblock_t new_pblk), \
+	TP_ARGS(ip, lblk, pblk, len, new_pblk))
+
+/* inode/irec events */
+DECLARE_EVENT_CLASS(xfs_inode_irec_class,
+	TP_PROTO(struct xfs_inode *ip, struct xfs_bmbt_irec *irec),
+	TP_ARGS(ip, irec),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_fileoff_t, lblk)
+		__field(xfs_extlen_t, len)
+		__field(xfs_fsblock_t, pblk)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->lblk = irec->br_startoff;
+		__entry->len = irec->br_blockcount;
+		__entry->pblk = irec->br_startblock;
+	),
+	TP_printk("dev %d:%d ino 0x%llx lblk 0x%llx len 0x%x pblk %llu",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->lblk,
+		  __entry->len,
+		  __entry->pblk)
+);
+#define DEFINE_INODE_IREC_EVENT(name) \
+DEFINE_EVENT(xfs_inode_irec_class, name, \
+	TP_PROTO(struct xfs_inode *ip, struct xfs_bmbt_irec *irec), \
+	TP_ARGS(ip, irec))
+
+/* refcount/reflink tracepoint definitions */
+
+/* reflink tracepoints */
+DEFINE_INODE_EVENT(xfs_reflink_set_inode_flag);
+DEFINE_INODE_EVENT(xfs_reflink_unset_inode_flag);
+DEFINE_ITRUNC_EVENT(xfs_reflink_update_inode_size);
+DEFINE_IOMAP_EVENT(xfs_reflink_remap_imap);
+TRACE_EVENT(xfs_reflink_remap_blocks_loop,
+	TP_PROTO(struct xfs_inode *src, xfs_fileoff_t soffset,
+		 xfs_filblks_t len, struct xfs_inode *dest,
+		 xfs_fileoff_t doffset),
+	TP_ARGS(src, soffset, len, dest, doffset),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, src_ino)
+		__field(xfs_fileoff_t, src_lblk)
+		__field(xfs_filblks_t, len)
+		__field(xfs_ino_t, dest_ino)
+		__field(xfs_fileoff_t, dest_lblk)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(src)->i_sb->s_dev;
+		__entry->src_ino = src->i_ino;
+		__entry->src_lblk = soffset;
+		__entry->len = len;
+		__entry->dest_ino = dest->i_ino;
+		__entry->dest_lblk = doffset;
+	),
+	TP_printk("dev %d:%d len 0x%llx "
+		  "ino 0x%llx offset 0x%llx blocks -> "
+		  "ino 0x%llx offset 0x%llx blocks",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->len,
+		  __entry->src_ino,
+		  __entry->src_lblk,
+		  __entry->dest_ino,
+		  __entry->dest_lblk)
+);
+TRACE_EVENT(xfs_reflink_punch_range,
+	TP_PROTO(struct xfs_inode *ip, xfs_fileoff_t lblk,
+		 xfs_extlen_t len),
+	TP_ARGS(ip, lblk, len),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_fileoff_t, lblk)
+		__field(xfs_extlen_t, len)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->lblk = lblk;
+		__entry->len = len;
+	),
+	TP_printk("dev %d:%d ino 0x%llx lblk 0x%llx len 0x%x",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->lblk,
+		  __entry->len)
+);
+TRACE_EVENT(xfs_reflink_remap,
+	TP_PROTO(struct xfs_inode *ip, xfs_fileoff_t lblk,
+		 xfs_extlen_t len, xfs_fsblock_t new_pblk),
+	TP_ARGS(ip, lblk, len, new_pblk),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_fileoff_t, lblk)
+		__field(xfs_extlen_t, len)
+		__field(xfs_fsblock_t, new_pblk)
+	),
+	TP_fast_assign(
+		__entry->dev = VFS_I(ip)->i_sb->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->lblk = lblk;
+		__entry->len = len;
+		__entry->new_pblk = new_pblk;
+	),
+	TP_printk("dev %d:%d ino 0x%llx lblk 0x%llx len 0x%x new_pblk %llu",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->lblk,
+		  __entry->len,
+		  __entry->new_pblk)
+);
+DEFINE_DOUBLE_IO_EVENT(xfs_reflink_remap_range);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_remap_range_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_set_inode_flag_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_update_inode_size_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_reflink_main_loop_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_read_iomap_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_remap_blocks_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_remap_extent_error);
+
+/* dedupe tracepoints */
+DEFINE_DOUBLE_IO_EVENT(xfs_reflink_compare_extents);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_compare_extents_error);
+
+/* ioctl tracepoints */
+DEFINE_DOUBLE_VFS_IO_EVENT(xfs_ioctl_reflink);
+DEFINE_DOUBLE_VFS_IO_EVENT(xfs_ioctl_clone_range);
+DEFINE_DOUBLE_VFS_IO_EVENT(xfs_ioctl_file_extent_same);
+TRACE_EVENT(xfs_ioctl_clone,
+	TP_PROTO(struct inode *src, struct inode *dest),
+	TP_ARGS(src, dest),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(unsigned long, src_ino)
+		__field(loff_t, src_isize)
+		__field(unsigned long, dest_ino)
+		__field(loff_t, dest_isize)
+	),
+	TP_fast_assign(
+		__entry->dev = src->i_sb->s_dev;
+		__entry->src_ino = src->i_ino;
+		__entry->src_isize = i_size_read(src);
+		__entry->dest_ino = dest->i_ino;
+		__entry->dest_isize = i_size_read(dest);
+	),
+	TP_printk("dev %d:%d "
+		  "ino 0x%lx isize 0x%llx -> "
+		  "ino 0x%lx isize 0x%llx\n",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->src_ino,
+		  __entry->src_isize,
+		  __entry->dest_ino,
+		  __entry->dest_isize)
+);
+
+/* unshare tracepoints */
+DEFINE_SIMPLE_IO_EVENT(xfs_reflink_unshare);
+DEFINE_SIMPLE_IO_EVENT(xfs_reflink_cow_eof_block);
+DEFINE_PAGE_EVENT(xfs_reflink_unshare_page);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_unshare_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_cow_eof_block_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_dirty_page_error);
+
+/* copy on write */
+DEFINE_INODE_IREC_EVENT(xfs_reflink_trim_around_shared);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_alloc);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_found);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_enospc);
+
+DEFINE_RW_EVENT(xfs_reflink_reserve_cow_range);
+DEFINE_RW_EVENT(xfs_reflink_allocate_cow_range);
+
+DEFINE_INODE_IREC_EVENT(xfs_reflink_bounce_dio_write);
+DEFINE_IOMAP_EVENT(xfs_reflink_find_cow_mapping);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_trim_irec);
+
+DEFINE_SIMPLE_IO_EVENT(xfs_reflink_cancel_cow_range);
+DEFINE_SIMPLE_IO_EVENT(xfs_reflink_end_cow);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_remap);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_remap_piece);
+
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_reserve_cow_range_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_allocate_cow_range_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_cancel_cow_range_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_end_cow_error);
+
+DEFINE_COW_EVENT(xfs_reflink_fork_buf);
+DEFINE_COW_EVENT(xfs_reflink_finish_fork_buf);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_fork_buf_error);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_finish_fork_buf_error);
+
+DEFINE_INODE_EVENT(xfs_reflink_cancel_pending_cow);
+DEFINE_INODE_IREC_EVENT(xfs_reflink_cancel_cow);
+DEFINE_INODE_ERROR_EVENT(xfs_reflink_cancel_pending_cow_error);
+
+/* rmap swapext tracepoints */
+DEFINE_INODE_IREC_EVENT(xfs_swap_extent_rmap_remap);
+DEFINE_INODE_IREC_EVENT(xfs_swap_extent_rmap_remap_piece);
+DEFINE_INODE_ERROR_EVENT(xfs_swap_extent_rmap_error);
+
 #endif /* _TRACE_XFS_H */
 
 #undef TRACE_INCLUDE_PATH

fs/xfs/xfs_trans.h

@@ -36,6 +36,11 @@ struct xfs_busy_extent;
 struct xfs_rud_log_item;
 struct xfs_rui_log_item;
 struct xfs_btree_cur;
+struct xfs_cui_log_item;
+struct xfs_cud_log_item;
+struct xfs_defer_ops;
+struct xfs_bui_log_item;
+struct xfs_bud_log_item;
 
 typedef struct xfs_log_item {
 	struct list_head		li_ail;		/* AIL pointers */
@@ -248,4 +253,28 @@ int xfs_trans_log_finish_rmap_update(struct xfs_trans *tp,
 		xfs_fsblock_t startblock, xfs_filblks_t blockcount,
 		xfs_exntst_t state, struct xfs_btree_cur **pcur);
 
+/* refcount updates */
+enum xfs_refcount_intent_type;
+
+void xfs_refcount_update_init_defer_op(void);
+struct xfs_cud_log_item *xfs_trans_get_cud(struct xfs_trans *tp,
+		struct xfs_cui_log_item *cuip);
+int xfs_trans_log_finish_refcount_update(struct xfs_trans *tp,
+		struct xfs_cud_log_item *cudp, struct xfs_defer_ops *dfops,
+		enum xfs_refcount_intent_type type, xfs_fsblock_t startblock,
+		xfs_extlen_t blockcount, xfs_fsblock_t *new_fsb,
+		xfs_extlen_t *new_len, struct xfs_btree_cur **pcur);
+
+/* mapping updates */
+enum xfs_bmap_intent_type;
+
+void xfs_bmap_update_init_defer_op(void);
+struct xfs_bud_log_item *xfs_trans_get_bud(struct xfs_trans *tp,
+		struct xfs_bui_log_item *buip);
+int xfs_trans_log_finish_bmap_update(struct xfs_trans *tp,
+		struct xfs_bud_log_item *rudp, struct xfs_defer_ops *dfops,
+		enum xfs_bmap_intent_type type, struct xfs_inode *ip,
+		int whichfork, xfs_fileoff_t startoff, xfs_fsblock_t startblock,
+		xfs_filblks_t blockcount, xfs_exntst_t state);
+
 #endif	/* __XFS_TRANS_H__ */

fs/xfs/xfs_trans_bmap.c

@@ -0,0 +1,249 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_trans.h"
+#include "xfs_trans_priv.h"
+#include "xfs_bmap_item.h"
+#include "xfs_alloc.h"
+#include "xfs_bmap.h"
+#include "xfs_inode.h"
+
+/*
+ * This routine is called to allocate a "bmap update done"
+ * log item.
+ */
+struct xfs_bud_log_item *
+xfs_trans_get_bud(
+	struct xfs_trans		*tp,
+	struct xfs_bui_log_item		*buip)
+{
+	struct xfs_bud_log_item		*budp;
+
+	budp = xfs_bud_init(tp->t_mountp, buip);
+	xfs_trans_add_item(tp, &budp->bud_item);
+	return budp;
+}
+
+/*
+ * Finish an bmap update and log it to the BUD. Note that the
+ * transaction is marked dirty regardless of whether the bmap update
+ * succeeds or fails to support the BUI/BUD lifecycle rules.
+ */
+int
+xfs_trans_log_finish_bmap_update(
+	struct xfs_trans		*tp,
+	struct xfs_bud_log_item		*budp,
+	struct xfs_defer_ops		*dop,
+	enum xfs_bmap_intent_type	type,
+	struct xfs_inode		*ip,
+	int				whichfork,
+	xfs_fileoff_t			startoff,
+	xfs_fsblock_t			startblock,
+	xfs_filblks_t			blockcount,
+	xfs_exntst_t			state)
+{
+	int				error;
+
+	error = xfs_bmap_finish_one(tp, dop, ip, type, whichfork, startoff,
+			startblock, blockcount, state);
+
+	/*
+	 * Mark the transaction dirty, even on error. This ensures the
+	 * transaction is aborted, which:
+	 *
+	 * 1.) releases the BUI and frees the BUD
+	 * 2.) shuts down the filesystem
+	 */
+	tp->t_flags |= XFS_TRANS_DIRTY;
+	budp->bud_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+	return error;
+}
+
+/* Sort bmap intents by inode. */
+static int
+xfs_bmap_update_diff_items(
+	void				*priv,
+	struct list_head		*a,
+	struct list_head		*b)
+{
+	struct xfs_bmap_intent		*ba;
+	struct xfs_bmap_intent		*bb;
+
+	ba = container_of(a, struct xfs_bmap_intent, bi_list);
+	bb = container_of(b, struct xfs_bmap_intent, bi_list);
+	return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
+}
+
+/* Get an BUI. */
+STATIC void *
+xfs_bmap_update_create_intent(
+	struct xfs_trans		*tp,
+	unsigned int			count)
+{
+	struct xfs_bui_log_item		*buip;
+
+	ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
+	ASSERT(tp != NULL);
+
+	buip = xfs_bui_init(tp->t_mountp);
+	ASSERT(buip != NULL);
+
+	/*
+	 * Get a log_item_desc to point at the new item.
+	 */
+	xfs_trans_add_item(tp, &buip->bui_item);
+	return buip;
+}
+
+/* Set the map extent flags for this mapping. */
+static void
+xfs_trans_set_bmap_flags(
+	struct xfs_map_extent		*bmap,
+	enum xfs_bmap_intent_type	type,
+	int				whichfork,
+	xfs_exntst_t			state)
+{
+	bmap->me_flags = 0;
+	switch (type) {
+	case XFS_BMAP_MAP:
+	case XFS_BMAP_UNMAP:
+		bmap->me_flags = type;
+		break;
+	default:
+		ASSERT(0);
+	}
+	if (state == XFS_EXT_UNWRITTEN)
+		bmap->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
+	if (whichfork == XFS_ATTR_FORK)
+		bmap->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
+}
+
+/* Log bmap updates in the intent item. */
+STATIC void
+xfs_bmap_update_log_item(
+	struct xfs_trans		*tp,
+	void				*intent,
+	struct list_head		*item)
+{
+	struct xfs_bui_log_item		*buip = intent;
+	struct xfs_bmap_intent		*bmap;
+	uint				next_extent;
+	struct xfs_map_extent		*map;
+
+	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
+
+	tp->t_flags |= XFS_TRANS_DIRTY;
+	buip->bui_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+	/*
+	 * atomic_inc_return gives us the value after the increment;
+	 * we want to use it as an array index so we need to subtract 1 from
+	 * it.
+	 */
+	next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
+	ASSERT(next_extent < buip->bui_format.bui_nextents);
+	map = &buip->bui_format.bui_extents[next_extent];
+	map->me_owner = bmap->bi_owner->i_ino;
+	map->me_startblock = bmap->bi_bmap.br_startblock;
+	map->me_startoff = bmap->bi_bmap.br_startoff;
+	map->me_len = bmap->bi_bmap.br_blockcount;
+	xfs_trans_set_bmap_flags(map, bmap->bi_type, bmap->bi_whichfork,
+			bmap->bi_bmap.br_state);
+}
+
+/* Get an BUD so we can process all the deferred rmap updates. */
+STATIC void *
+xfs_bmap_update_create_done(
+	struct xfs_trans		*tp,
+	void				*intent,
+	unsigned int			count)
+{
+	return xfs_trans_get_bud(tp, intent);
+}
+
+/* Process a deferred rmap update. */
+STATIC int
+xfs_bmap_update_finish_item(
+	struct xfs_trans		*tp,
+	struct xfs_defer_ops		*dop,
+	struct list_head		*item,
+	void				*done_item,
+	void				**state)
+{
+	struct xfs_bmap_intent		*bmap;
+	int				error;
+
+	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
+	error = xfs_trans_log_finish_bmap_update(tp, done_item, dop,
+			bmap->bi_type,
+			bmap->bi_owner, bmap->bi_whichfork,
+			bmap->bi_bmap.br_startoff,
+			bmap->bi_bmap.br_startblock,
+			bmap->bi_bmap.br_blockcount,
+			bmap->bi_bmap.br_state);
+	kmem_free(bmap);
+	return error;
+}
+
+/* Abort all pending BUIs. */
+STATIC void
+xfs_bmap_update_abort_intent(
+	void				*intent)
+{
+	xfs_bui_release(intent);
+}
+
+/* Cancel a deferred rmap update. */
+STATIC void
+xfs_bmap_update_cancel_item(
+	struct list_head		*item)
+{
+	struct xfs_bmap_intent		*bmap;
+
+	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
+	kmem_free(bmap);
+}
+
+static const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
+	.type		= XFS_DEFER_OPS_TYPE_BMAP,
+	.max_items	= XFS_BUI_MAX_FAST_EXTENTS,
+	.diff_items	= xfs_bmap_update_diff_items,
+	.create_intent	= xfs_bmap_update_create_intent,
+	.abort_intent	= xfs_bmap_update_abort_intent,
+	.log_item	= xfs_bmap_update_log_item,
+	.create_done	= xfs_bmap_update_create_done,
+	.finish_item	= xfs_bmap_update_finish_item,
+	.cancel_item	= xfs_bmap_update_cancel_item,
+};
+
+/* Register the deferred op type. */
+void
+xfs_bmap_update_init_defer_op(void)
+{
+	xfs_defer_init_op_type(&xfs_bmap_update_defer_type);
+}

fs/xfs/xfs_trans_refcount.c

@@ -0,0 +1,264 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_trans.h"
+#include "xfs_trans_priv.h"
+#include "xfs_refcount_item.h"
+#include "xfs_alloc.h"
+#include "xfs_refcount.h"
+
+/*
+ * This routine is called to allocate a "refcount update done"
+ * log item.
+ */
+struct xfs_cud_log_item *
+xfs_trans_get_cud(
+	struct xfs_trans		*tp,
+	struct xfs_cui_log_item		*cuip)
+{
+	struct xfs_cud_log_item		*cudp;
+
+	cudp = xfs_cud_init(tp->t_mountp, cuip);
+	xfs_trans_add_item(tp, &cudp->cud_item);
+	return cudp;
+}
+
+/*
+ * Finish an refcount update and log it to the CUD. Note that the
+ * transaction is marked dirty regardless of whether the refcount
+ * update succeeds or fails to support the CUI/CUD lifecycle rules.
+ */
+int
+xfs_trans_log_finish_refcount_update(
+	struct xfs_trans		*tp,
+	struct xfs_cud_log_item		*cudp,
+	struct xfs_defer_ops		*dop,
+	enum xfs_refcount_intent_type	type,
+	xfs_fsblock_t			startblock,
+	xfs_extlen_t			blockcount,
+	xfs_fsblock_t			*new_fsb,
+	xfs_extlen_t			*new_len,
+	struct xfs_btree_cur		**pcur)
+{
+	int				error;
+
+	error = xfs_refcount_finish_one(tp, dop, type, startblock,
+			blockcount, new_fsb, new_len, pcur);
+
+	/*
+	 * Mark the transaction dirty, even on error. This ensures the
+	 * transaction is aborted, which:
+	 *
+	 * 1.) releases the CUI and frees the CUD
+	 * 2.) shuts down the filesystem
+	 */
+	tp->t_flags |= XFS_TRANS_DIRTY;
+	cudp->cud_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+	return error;
+}
+
+/* Sort refcount intents by AG. */
+static int
+xfs_refcount_update_diff_items(
+	void				*priv,
+	struct list_head		*a,
+	struct list_head		*b)
+{
+	struct xfs_mount		*mp = priv;
+	struct xfs_refcount_intent	*ra;
+	struct xfs_refcount_intent	*rb;
+
+	ra = container_of(a, struct xfs_refcount_intent, ri_list);
+	rb = container_of(b, struct xfs_refcount_intent, ri_list);
+	return  XFS_FSB_TO_AGNO(mp, ra->ri_startblock) -
+		XFS_FSB_TO_AGNO(mp, rb->ri_startblock);
+}
+
+/* Get an CUI. */
+STATIC void *
+xfs_refcount_update_create_intent(
+	struct xfs_trans		*tp,
+	unsigned int			count)
+{
+	struct xfs_cui_log_item		*cuip;
+
+	ASSERT(tp != NULL);
+	ASSERT(count > 0);
+
+	cuip = xfs_cui_init(tp->t_mountp, count);
+	ASSERT(cuip != NULL);
+
+	/*
+	 * Get a log_item_desc to point at the new item.
+	 */
+	xfs_trans_add_item(tp, &cuip->cui_item);
+	return cuip;
+}
+
+/* Set the phys extent flags for this reverse mapping. */
+static void
+xfs_trans_set_refcount_flags(
+	struct xfs_phys_extent		*refc,
+	enum xfs_refcount_intent_type	type)
+{
+	refc->pe_flags = 0;
+	switch (type) {
+	case XFS_REFCOUNT_INCREASE:
+	case XFS_REFCOUNT_DECREASE:
+	case XFS_REFCOUNT_ALLOC_COW:
+	case XFS_REFCOUNT_FREE_COW:
+		refc->pe_flags |= type;
+		break;
+	default:
+		ASSERT(0);
+	}
+}
+
+/* Log refcount updates in the intent item. */
+STATIC void
+xfs_refcount_update_log_item(
+	struct xfs_trans		*tp,
+	void				*intent,
+	struct list_head		*item)
+{
+	struct xfs_cui_log_item		*cuip = intent;
+	struct xfs_refcount_intent	*refc;
+	uint				next_extent;
+	struct xfs_phys_extent		*ext;
+
+	refc = container_of(item, struct xfs_refcount_intent, ri_list);
+
+	tp->t_flags |= XFS_TRANS_DIRTY;
+	cuip->cui_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+	/*
+	 * atomic_inc_return gives us the value after the increment;
+	 * we want to use it as an array index so we need to subtract 1 from
+	 * it.
+	 */
+	next_extent = atomic_inc_return(&cuip->cui_next_extent) - 1;
+	ASSERT(next_extent < cuip->cui_format.cui_nextents);
+	ext = &cuip->cui_format.cui_extents[next_extent];
+	ext->pe_startblock = refc->ri_startblock;
+	ext->pe_len = refc->ri_blockcount;
+	xfs_trans_set_refcount_flags(ext, refc->ri_type);
+}
+
+/* Get an CUD so we can process all the deferred refcount updates. */
+STATIC void *
+xfs_refcount_update_create_done(
+	struct xfs_trans		*tp,
+	void				*intent,
+	unsigned int			count)
+{
+	return xfs_trans_get_cud(tp, intent);
+}
+
+/* Process a deferred refcount update. */
+STATIC int
+xfs_refcount_update_finish_item(
+	struct xfs_trans		*tp,
+	struct xfs_defer_ops		*dop,
+	struct list_head		*item,
+	void				*done_item,
+	void				**state)
+{
+	struct xfs_refcount_intent	*refc;
+	xfs_fsblock_t			new_fsb;
+	xfs_extlen_t			new_aglen;
+	int				error;
+
+	refc = container_of(item, struct xfs_refcount_intent, ri_list);
+	error = xfs_trans_log_finish_refcount_update(tp, done_item, dop,
+			refc->ri_type,
+			refc->ri_startblock,
+			refc->ri_blockcount,
+			&new_fsb, &new_aglen,
+			(struct xfs_btree_cur **)state);
+	/* Did we run out of reservation?  Requeue what we didn't finish. */
+	if (!error && new_aglen > 0) {
+		ASSERT(refc->ri_type == XFS_REFCOUNT_INCREASE ||
+		       refc->ri_type == XFS_REFCOUNT_DECREASE);
+		refc->ri_startblock = new_fsb;
+		refc->ri_blockcount = new_aglen;
+		return -EAGAIN;
+	}
+	kmem_free(refc);
+	return error;
+}
+
+/* Clean up after processing deferred refcounts. */
+STATIC void
+xfs_refcount_update_finish_cleanup(
+	struct xfs_trans	*tp,
+	void			*state,
+	int			error)
+{
+	struct xfs_btree_cur	*rcur = state;
+
+	xfs_refcount_finish_one_cleanup(tp, rcur, error);
+}
+
+/* Abort all pending CUIs. */
+STATIC void
+xfs_refcount_update_abort_intent(
+	void				*intent)
+{
+	xfs_cui_release(intent);
+}
+
+/* Cancel a deferred refcount update. */
+STATIC void
+xfs_refcount_update_cancel_item(
+	struct list_head		*item)
+{
+	struct xfs_refcount_intent	*refc;
+
+	refc = container_of(item, struct xfs_refcount_intent, ri_list);
+	kmem_free(refc);
+}
+
+static const struct xfs_defer_op_type xfs_refcount_update_defer_type = {
+	.type		= XFS_DEFER_OPS_TYPE_REFCOUNT,
+	.max_items	= XFS_CUI_MAX_FAST_EXTENTS,
+	.diff_items	= xfs_refcount_update_diff_items,
+	.create_intent	= xfs_refcount_update_create_intent,
+	.abort_intent	= xfs_refcount_update_abort_intent,
+	.log_item	= xfs_refcount_update_log_item,
+	.create_done	= xfs_refcount_update_create_done,
+	.finish_item	= xfs_refcount_update_finish_item,
+	.finish_cleanup = xfs_refcount_update_finish_cleanup,
+	.cancel_item	= xfs_refcount_update_cancel_item,
+};
+
+/* Register the deferred op type. */
+void
+xfs_refcount_update_init_defer_op(void)
+{
+	xfs_defer_init_op_type(&xfs_refcount_update_defer_type);
+}

fs/xfs/xfs_trans_rmap.c

@@ -48,12 +48,21 @@ xfs_trans_set_rmap_flags(
 	case XFS_RMAP_MAP:
 		rmap->me_flags |= XFS_RMAP_EXTENT_MAP;
 		break;
+	case XFS_RMAP_MAP_SHARED:
+		rmap->me_flags |= XFS_RMAP_EXTENT_MAP_SHARED;
+		break;
 	case XFS_RMAP_UNMAP:
 		rmap->me_flags |= XFS_RMAP_EXTENT_UNMAP;
 		break;
+	case XFS_RMAP_UNMAP_SHARED:
+		rmap->me_flags |= XFS_RMAP_EXTENT_UNMAP_SHARED;
+		break;
 	case XFS_RMAP_CONVERT:
 		rmap->me_flags |= XFS_RMAP_EXTENT_CONVERT;
 		break;
+	case XFS_RMAP_CONVERT_SHARED:
+		rmap->me_flags |= XFS_RMAP_EXTENT_CONVERT_SHARED;
+		break;
 	case XFS_RMAP_ALLOC:
 		rmap->me_flags |= XFS_RMAP_EXTENT_ALLOC;
 		break;

include/linux/falloc.h

@@ -25,6 +25,7 @@ struct space_resv {
 					 FALLOC_FL_PUNCH_HOLE |		\
 					 FALLOC_FL_COLLAPSE_RANGE |	\
 					 FALLOC_FL_ZERO_RANGE |		\
-					 FALLOC_FL_INSERT_RANGE)
+					 FALLOC_FL_INSERT_RANGE |	\
+					 FALLOC_FL_UNSHARE_RANGE)
 
 #endif /* _FALLOC_H_ */

include/uapi/linux/falloc.h

@@ -58,4 +58,22 @@
  */
 #define FALLOC_FL_INSERT_RANGE		0x20
 
+/*
+ * FALLOC_FL_UNSHARE_RANGE is used to unshare shared blocks within the
+ * file size without overwriting any existing data. The purpose of this
+ * call is to preemptively reallocate any blocks that are subject to
+ * copy-on-write.
+ *
+ * Different filesystems may implement different limitations on the
+ * granularity of the operation. Most will limit operations to filesystem
+ * block size boundaries, but this boundary may be larger or smaller
+ * depending on the filesystem and/or the configuration of the filesystem
+ * or file.
+ *
+ * This flag can only be used with allocate-mode fallocate, which is
+ * to say that it cannot be used with the punch, zero, collapse, or
+ * insert range modes.
+ */
+#define FALLOC_FL_UNSHARE_RANGE		0x40
+
 #endif /* _UAPI_FALLOC_H_ */

include/uapi/linux/fs.h

@@ -158,7 +158,8 @@ struct fsxattr {
 	__u32		fsx_extsize;	/* extsize field value (get/set)*/
 	__u32		fsx_nextents;	/* nextents field value (get)	*/
 	__u32		fsx_projid;	/* project identifier (get/set) */
-	unsigned char	fsx_pad[12];
+	__u32		fsx_cowextsize;	/* CoW extsize field value (get/set)*/
+	unsigned char	fsx_pad[8];
 };
 
 /*
@@ -179,6 +180,7 @@ struct fsxattr {
 #define FS_XFLAG_NODEFRAG	0x00002000	/* do not defragment */
 #define FS_XFLAG_FILESTREAM	0x00004000	/* use filestream allocator */
 #define FS_XFLAG_DAX		0x00008000	/* use DAX for IO */
+#define FS_XFLAG_COWEXTSIZE	0x00010000	/* CoW extent size allocator hint */
 #define FS_XFLAG_HASATTR	0x80000000	/* no DIFLAG for this	*/
 
 /* the read-only stuff doesn't really belong here, but any other place is