Merge tag 'xfs-for-linus-4.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs

Pul xfs updates from Dave Chinner:
 "There's a couple of small API changes to the core DAX code which
  required small changes to the ext2 and ext4 code bases, but otherwise
  everything is within the XFS codebase.

  This update contains:

   - A new sparse on-disk inode record format to allow small extents to
     be used for inode allocation when free space is fragmented.

   - DAX support.  This includes minor changes to the DAX core code to
     fix problems with lock ordering and bufferhead mapping abuse.

   - transaction commit interface cleanup

   - removal of various unnecessary XFS specific type definitions

   - cleanup and optimisation of freelist preparation before allocation

   - various minor cleanups

   - bug fixes for
	- transaction reservation leaks
	- incorrect inode logging in unwritten extent conversion
	- mmap lock vs freeze ordering
	- remote symlink mishandling
	- attribute fork removal issues"

* tag 'xfs-for-linus-4.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (49 commits)
  xfs: don't truncate attribute extents if no extents exist
  xfs: clean up XFS_MIN_FREELIST macros
  xfs: sanitise error handling in xfs_alloc_fix_freelist
  xfs: factor out free space extent length check
  xfs: xfs_alloc_fix_freelist() can use incore perag structures
  xfs: remove xfs_caddr_t
  xfs: use void pointers in log validation helpers
  xfs: return a void pointer from xfs_buf_offset
  xfs: remove inst_t
  xfs: remove __psint_t and __psunsigned_t
  xfs: fix remote symlinks on V5/CRC filesystems
  xfs: fix xfs_log_done interface
  xfs: saner xfs_trans_commit interface
  xfs: remove the flags argument to xfs_trans_cancel
  xfs: pass a boolean flag to xfs_trans_free_items
  xfs: switch remaining xfs_trans_dup users to xfs_trans_roll
  xfs: check min blks for random debug mode sparse allocations
  xfs: fix sparse inodes 32-bit compile failure
  xfs: add initial DAX support
  xfs: add DAX IO path support
  ...

15 files changed, 889 insertions, 236 deletions

diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
index 516162be1398..f9e9ffe6fb46 100644
--- a/fs/xfs/libxfs/xfs_alloc.c
+++ b/fs/xfs/libxfs/xfs_alloc.c
@@ -149,13 +149,27 @@ xfs_alloc_compute_aligned(
 {
 	xfs_agblock_t	bno;
 	xfs_extlen_t	len;
+	xfs_extlen_t	diff;
 
 	/* Trim busy sections out of found extent */
 	xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len);
 
+	/*
+	 * If we have a largish extent that happens to start before min_agbno,
+	 * see if we can shift it into range...
+	 */
+	if (bno < args->min_agbno && bno + len > args->min_agbno) {
+		diff = args->min_agbno - bno;
+		if (len > diff) {
+			bno += diff;
+			len -= diff;
+		}
+	}
+
 	if (args->alignment > 1 && len >= args->minlen) {
 		xfs_agblock_t	aligned_bno = roundup(bno, args->alignment);
-		xfs_extlen_t	diff = aligned_bno - bno;
+
+		diff = aligned_bno - bno;
 
 		*resbno = aligned_bno;
 		*reslen = diff >= len ? 0 : len - diff;
@@ -795,9 +809,13 @@ xfs_alloc_find_best_extent(
 		 * The good extent is closer than this one.
 		 */
 		if (!dir) {
+			if (*sbnoa > args->max_agbno)
+				goto out_use_good;
 			if (*sbnoa >= args->agbno + gdiff)
 				goto out_use_good;
 		} else {
+			if (*sbnoa < args->min_agbno)
+				goto out_use_good;
 			if (*sbnoa <= args->agbno - gdiff)
 				goto out_use_good;
 		}
@@ -884,6 +902,17 @@ xfs_alloc_ag_vextent_near(
 	dofirst = prandom_u32() & 1;
 #endif
 
+	/* handle unitialized agbno range so caller doesn't have to */
+	if (!args->min_agbno && !args->max_agbno)
+		args->max_agbno = args->mp->m_sb.sb_agblocks - 1;
+	ASSERT(args->min_agbno <= args->max_agbno);
+
+	/* clamp agbno to the range if it's outside */
+	if (args->agbno < args->min_agbno)
+		args->agbno = args->min_agbno;
+	if (args->agbno > args->max_agbno)
+		args->agbno = args->max_agbno;
+
 restart:
 	bno_cur_lt = NULL;
 	bno_cur_gt = NULL;
@@ -976,6 +1005,8 @@ restart:
 						  &ltbnoa, &ltlena);
 			if (ltlena < args->minlen)
 				continue;
+			if (ltbnoa < args->min_agbno || ltbnoa > args->max_agbno)
+				continue;
 			args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
 			xfs_alloc_fix_len(args);
 			ASSERT(args->len >= args->minlen);
@@ -1096,11 +1127,11 @@ restart:
 			XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
 			xfs_alloc_compute_aligned(args, ltbno, ltlen,
 						  &ltbnoa, &ltlena);
-			if (ltlena >= args->minlen)
+			if (ltlena >= args->minlen && ltbnoa >= args->min_agbno)
 				break;
 			if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
 				goto error0;
-			if (!i) {
+			if (!i || ltbnoa < args->min_agbno) {
 				xfs_btree_del_cursor(bno_cur_lt,
 						     XFS_BTREE_NOERROR);
 				bno_cur_lt = NULL;
@@ -1112,11 +1143,11 @@ restart:
 			XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
 			xfs_alloc_compute_aligned(args, gtbno, gtlen,
 						  &gtbnoa, &gtlena);
-			if (gtlena >= args->minlen)
+			if (gtlena >= args->minlen && gtbnoa <= args->max_agbno)
 				break;
 			if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
 				goto error0;
-			if (!i) {
+			if (!i || gtbnoa > args->max_agbno) {
 				xfs_btree_del_cursor(bno_cur_gt,
 						     XFS_BTREE_NOERROR);
 				bno_cur_gt = NULL;
@@ -1216,6 +1247,7 @@ restart:
 	ASSERT(ltnew >= ltbno);
 	ASSERT(ltnew + rlen <= ltbnoa + ltlena);
 	ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
+	ASSERT(ltnew >= args->min_agbno && ltnew <= args->max_agbno);
 	args->agbno = ltnew;
 
 	if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
@@ -1825,11 +1857,11 @@ xfs_alloc_compute_maxlevels(
 xfs_extlen_t
 xfs_alloc_longest_free_extent(
 	struct xfs_mount	*mp,
-	struct xfs_perag	*pag)
+	struct xfs_perag	*pag,
+	xfs_extlen_t		need)
 {
-	xfs_extlen_t		need, delta = 0;
+	xfs_extlen_t		delta = 0;
 
-	need = XFS_MIN_FREELIST_PAG(pag, mp);
 	if (need > pag->pagf_flcount)
 		delta = need - pag->pagf_flcount;
 
@@ -1838,131 +1870,150 @@ xfs_alloc_longest_free_extent(
 	return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
 }
 
+unsigned int
+xfs_alloc_min_freelist(
+	struct xfs_mount	*mp,
+	struct xfs_perag	*pag)
+{
+	unsigned int		min_free;
+
+	/* space needed by-bno freespace btree */
+	min_free = min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_BNOi] + 1,
+				       mp->m_ag_maxlevels);
+	/* space needed by-size freespace btree */
+	min_free += min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_CNTi] + 1,
+				       mp->m_ag_maxlevels);
+
+	return min_free;
+}
+
+/*
+ * Check if the operation we are fixing up the freelist for should go ahead or
+ * not. If we are freeing blocks, we always allow it, otherwise the allocation
+ * is dependent on whether the size and shape of free space available will
+ * permit the requested allocation to take place.
+ */
+static bool
+xfs_alloc_space_available(
+	struct xfs_alloc_arg	*args,
+	xfs_extlen_t		min_free,
+	int			flags)
+{
+	struct xfs_perag	*pag = args->pag;
+	xfs_extlen_t		longest;
+	int			available;
+
+	if (flags & XFS_ALLOC_FLAG_FREEING)
+		return true;
+
+	/* do we have enough contiguous free space for the allocation? */
+	longest = xfs_alloc_longest_free_extent(args->mp, pag, min_free);
+	if ((args->minlen + args->alignment + args->minalignslop - 1) > longest)
+		return false;
+
+	/* do have enough free space remaining for the allocation? */
+	available = (int)(pag->pagf_freeblks + pag->pagf_flcount -
+			  min_free - args->total);
+	if (available < (int)args->minleft)
+		return false;
+
+	return true;
+}
+
 /*
  * Decide whether to use this allocation group for this allocation.
  * If so, fix up the btree freelist's size.
  */
 STATIC int			/* error */
 xfs_alloc_fix_freelist(
-	xfs_alloc_arg_t	*args,	/* allocation argument structure */
-	int		flags)	/* XFS_ALLOC_FLAG_... */
+	struct xfs_alloc_arg	*args,	/* allocation argument structure */
+	int			flags)	/* XFS_ALLOC_FLAG_... */
 {
-	xfs_buf_t	*agbp;	/* agf buffer pointer */
-	xfs_agf_t	*agf;	/* a.g. freespace structure pointer */
-	xfs_buf_t	*agflbp;/* agfl buffer pointer */
-	xfs_agblock_t	bno;	/* freelist block */
-	xfs_extlen_t	delta;	/* new blocks needed in freelist */
-	int		error;	/* error result code */
-	xfs_extlen_t	longest;/* longest extent in allocation group */
-	xfs_mount_t	*mp;	/* file system mount point structure */
-	xfs_extlen_t	need;	/* total blocks needed in freelist */
-	xfs_perag_t	*pag;	/* per-ag information structure */
-	xfs_alloc_arg_t	targs;	/* local allocation arguments */
-	xfs_trans_t	*tp;	/* transaction pointer */
-
-	mp = args->mp;
+	struct xfs_mount	*mp = args->mp;
+	struct xfs_perag	*pag = args->pag;
+	struct xfs_trans	*tp = args->tp;
+	struct xfs_buf		*agbp = NULL;
+	struct xfs_buf		*agflbp = NULL;
+	struct xfs_alloc_arg	targs;	/* local allocation arguments */
+	xfs_agblock_t		bno;	/* freelist block */
+	xfs_extlen_t		need;	/* total blocks needed in freelist */
+	int			error;
 
-	pag = args->pag;
-	tp = args->tp;
 	if (!pag->pagf_init) {
-		if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
-				&agbp)))
-			return error;
+		error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
+		if (error)
+			goto out_no_agbp;
 		if (!pag->pagf_init) {
 			ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
 			ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
-			args->agbp = NULL;
-			return 0;
+			goto out_agbp_relse;
 		}
-	} else
-		agbp = NULL;
+	}
 
 	/*
-	 * If this is a metadata preferred pag and we are user data
-	 * then try somewhere else if we are not being asked to
-	 * try harder at this point
+	 * If this is a metadata preferred pag and we are user data then try
+	 * somewhere else if we are not being asked to try harder at this
+	 * point
 	 */
 	if (pag->pagf_metadata && args->userdata &&
 	    (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
 		ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
-		args->agbp = NULL;
-		return 0;
+		goto out_agbp_relse;
 	}
 
-	if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
-		/*
-		 * If it looks like there isn't a long enough extent, or enough
-		 * total blocks, reject it.
-		 */
-		need = XFS_MIN_FREELIST_PAG(pag, mp);
-		longest = xfs_alloc_longest_free_extent(mp, pag);
-		if ((args->minlen + args->alignment + args->minalignslop - 1) >
-				longest ||
-		    ((int)(pag->pagf_freeblks + pag->pagf_flcount -
-			   need - args->total) < (int)args->minleft)) {
-			if (agbp)
-				xfs_trans_brelse(tp, agbp);
-			args->agbp = NULL;
-			return 0;
-		}
-	}
+	need = xfs_alloc_min_freelist(mp, pag);
+	if (!xfs_alloc_space_available(args, need, flags))
+		goto out_agbp_relse;
 
 	/*
 	 * Get the a.g. freespace buffer.
 	 * Can fail if we're not blocking on locks, and it's held.
 	 */
-	if (agbp == NULL) {
-		if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
-				&agbp)))
-			return error;
-		if (agbp == NULL) {
+	if (!agbp) {
+		error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
+		if (error)
+			goto out_no_agbp;
+		if (!agbp) {
 			ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
 			ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
-			args->agbp = NULL;
-			return 0;
-		}
-	}
-	/*
-	 * Figure out how many blocks we should have in the freelist.
-	 */
-	agf = XFS_BUF_TO_AGF(agbp);
-	need = XFS_MIN_FREELIST(agf, mp);
-	/*
-	 * If there isn't enough total or single-extent, reject it.
-	 */
-	if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
-		delta = need > be32_to_cpu(agf->agf_flcount) ?
-			(need - be32_to_cpu(agf->agf_flcount)) : 0;
-		longest = be32_to_cpu(agf->agf_longest);
-		longest = (longest > delta) ? (longest - delta) :
-			(be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
-		if ((args->minlen + args->alignment + args->minalignslop - 1) >
-				longest ||
-		    ((int)(be32_to_cpu(agf->agf_freeblks) +
-		     be32_to_cpu(agf->agf_flcount) - need - args->total) <
-				(int)args->minleft)) {
-			xfs_trans_brelse(tp, agbp);
-			args->agbp = NULL;
-			return 0;
+			goto out_no_agbp;
 		}
 	}
+
+	/* If there isn't enough total space or single-extent, reject it. */
+	need = xfs_alloc_min_freelist(mp, pag);
+	if (!xfs_alloc_space_available(args, need, flags))
+		goto out_agbp_relse;
+
 	/*
 	 * Make the freelist shorter if it's too long.
+	 *
+	 * Note that from this point onwards, we will always release the agf and
+	 * agfl buffers on error. This handles the case where we error out and
+	 * the buffers are clean or may not have been joined to the transaction
+	 * and hence need to be released manually. If they have been joined to
+	 * the transaction, then xfs_trans_brelse() will handle them
+	 * appropriately based on the recursion count and dirty state of the
+	 * buffer.
+	 *
+	 * XXX (dgc): When we have lots of free space, does this buy us
+	 * anything other than extra overhead when we need to put more blocks
+	 * back on the free list? Maybe we should only do this when space is
+	 * getting low or the AGFL is more than half full?
 	 */
-	while (be32_to_cpu(agf->agf_flcount) > need) {
-		xfs_buf_t	*bp;
+	while (pag->pagf_flcount > need) {
+		struct xfs_buf	*bp;
 
 		error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
 		if (error)
-			return error;
-		if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
-			return error;
+			goto out_agbp_relse;
+		error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1);
+		if (error)
+			goto out_agbp_relse;
 		bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
 		xfs_trans_binval(tp, bp);
 	}
-	/*
-	 * Initialize the args structure.
-	 */
+
 	memset(&targs, 0, sizeof(targs));
 	targs.tp = tp;
 	targs.mp = mp;
@@ -1971,21 +2022,20 @@ xfs_alloc_fix_freelist(
 	targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
 	targs.type = XFS_ALLOCTYPE_THIS_AG;
 	targs.pag = pag;
-	if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
-		return error;
-	/*
-	 * Make the freelist longer if it's too short.
-	 */
-	while (be32_to_cpu(agf->agf_flcount) < need) {
+	error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp);
+	if (error)
+		goto out_agbp_relse;
+
+	/* Make the freelist longer if it's too short. */
+	while (pag->pagf_flcount < need) {
 		targs.agbno = 0;
-		targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
-		/*
-		 * Allocate as many blocks as possible at once.
-		 */
-		if ((error = xfs_alloc_ag_vextent(&targs))) {
-			xfs_trans_brelse(tp, agflbp);
-			return error;
-		}
+		targs.maxlen = need - pag->pagf_flcount;
+
+		/* Allocate as many blocks as possible at once. */
+		error = xfs_alloc_ag_vextent(&targs);
+		if (error)
+			goto out_agflbp_relse;
+
 		/*
 		 * Stop if we run out.  Won't happen if callers are obeying
 		 * the restrictions correctly.  Can happen for free calls
@@ -1994,9 +2044,7 @@ xfs_alloc_fix_freelist(
 		if (targs.agbno == NULLAGBLOCK) {
 			if (flags & XFS_ALLOC_FLAG_FREEING)
 				break;
-			xfs_trans_brelse(tp, agflbp);
-			args->agbp = NULL;
-			return 0;
+			goto out_agflbp_relse;
 		}
 		/*
 		 * Put each allocated block on the list.
@@ -2005,12 +2053,21 @@ xfs_alloc_fix_freelist(
 			error = xfs_alloc_put_freelist(tp, agbp,
 							agflbp, bno, 0);
 			if (error)
-				return error;
+				goto out_agflbp_relse;
 		}
 	}
 	xfs_trans_brelse(tp, agflbp);
 	args->agbp = agbp;
 	return 0;
+
+out_agflbp_relse:
+	xfs_trans_brelse(tp, agflbp);
+out_agbp_relse:
+	if (agbp)
+		xfs_trans_brelse(tp, agbp);
+out_no_agbp:
+	args->agbp = NULL;
+	return error;
 }
 
 /*
diff --git a/fs/xfs/libxfs/xfs_alloc.h b/fs/xfs/libxfs/xfs_alloc.h
index d1b4b6a5c894..ca1c8168373a 100644
--- a/fs/xfs/libxfs/xfs_alloc.h
+++ b/fs/xfs/libxfs/xfs_alloc.h
@@ -112,6 +112,8 @@ typedef struct xfs_alloc_arg {
 	xfs_extlen_t	total;		/* total blocks needed in xaction */
 	xfs_extlen_t	alignment;	/* align answer to multiple of this */
 	xfs_extlen_t	minalignslop;	/* slop for minlen+alignment calcs */
+	xfs_agblock_t	min_agbno;	/* set an agbno range for NEAR allocs */
+	xfs_agblock_t	max_agbno;	/* ... */
 	xfs_extlen_t	len;		/* output: actual size of extent */
 	xfs_alloctype_t	type;		/* allocation type XFS_ALLOCTYPE_... */
 	xfs_alloctype_t	otype;		/* original allocation type */
@@ -128,11 +130,9 @@ typedef struct xfs_alloc_arg {
 #define XFS_ALLOC_USERDATA		1	/* allocation is for user data*/
 #define XFS_ALLOC_INITIAL_USER_DATA	2	/* special case start of file */
 
-/*
- * Find the length of the longest extent in an AG.
- */
-xfs_extlen_t
-xfs_alloc_longest_free_extent(struct xfs_mount *mp,
+xfs_extlen_t xfs_alloc_longest_free_extent(struct xfs_mount *mp,
+		struct xfs_perag *pag, xfs_extlen_t need);
+unsigned int xfs_alloc_min_freelist(struct xfs_mount *mp,
 		struct xfs_perag *pag);
 
 /*
diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
index 0a472fbe06d4..3349c9a1e845 100644
--- a/fs/xfs/libxfs/xfs_attr.c
+++ b/fs/xfs/libxfs/xfs_attr.c
@@ -266,7 +266,7 @@ xfs_attr_set(
 	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
 	error = xfs_trans_reserve(args.trans, &tres, args.total, 0);
 	if (error) {
-		xfs_trans_cancel(args.trans, 0);
+		xfs_trans_cancel(args.trans);
 		return error;
 	}
 	xfs_ilock(dp, XFS_ILOCK_EXCL);
@@ -276,7 +276,7 @@ xfs_attr_set(
 				       XFS_QMOPT_RES_REGBLKS);
 	if (error) {
 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
-		xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
+		xfs_trans_cancel(args.trans);
 		return error;
 	}
 
@@ -320,8 +320,7 @@ xfs_attr_set(
 				xfs_trans_ichgtime(args.trans, dp,
 							XFS_ICHGTIME_CHG);
 			}
-			err2 = xfs_trans_commit(args.trans,
-						 XFS_TRANS_RELEASE_LOG_RES);
+			err2 = xfs_trans_commit(args.trans);
 			xfs_iunlock(dp, XFS_ILOCK_EXCL);
 
 			return error ? error : err2;
@@ -383,16 +382,14 @@ xfs_attr_set(
 	 * Commit the last in the sequence of transactions.
 	 */
 	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
-	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
+	error = xfs_trans_commit(args.trans);
 	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 
 	return error;
 
 out:
-	if (args.trans) {
-		xfs_trans_cancel(args.trans,
-			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
-	}
+	if (args.trans)
+		xfs_trans_cancel(args.trans);
 	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 	return error;
 }
@@ -462,7 +459,7 @@ xfs_attr_remove(
 	error = xfs_trans_reserve(args.trans, &M_RES(mp)->tr_attrrm,
 				  XFS_ATTRRM_SPACE_RES(mp), 0);
 	if (error) {
-		xfs_trans_cancel(args.trans, 0);
+		xfs_trans_cancel(args.trans);
 		return error;
 	}
 
@@ -501,16 +498,14 @@ xfs_attr_remove(
 	 * Commit the last in the sequence of transactions.
 	 */
 	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
-	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
+	error = xfs_trans_commit(args.trans);
 	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 
 	return error;
 
 out:
-	if (args.trans) {
-		xfs_trans_cancel(args.trans,
-			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
-	}
+	if (args.trans)
+		xfs_trans_cancel(args.trans);
 	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 	return error;
 }
diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
index f1026e86dabc..63e05b663380 100644
--- a/fs/xfs/libxfs/xfs_bmap.c
+++ b/fs/xfs/libxfs/xfs_bmap.c
@@ -1112,7 +1112,6 @@ xfs_bmap_add_attrfork(
 	int			committed;	/* xaction was committed */
 	int			logflags;	/* logging flags */
 	int			error;		/* error return value */
-	int			cancel_flags = 0;
 
 	ASSERT(XFS_IFORK_Q(ip) == 0);
 
@@ -1124,17 +1123,15 @@ xfs_bmap_add_attrfork(
 		tp->t_flags |= XFS_TRANS_RESERVE;
 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
 	if (error) {
-		xfs_trans_cancel(tp, 0);
+		xfs_trans_cancel(tp);
 		return error;
 	}
-	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
 	xfs_ilock(ip, XFS_ILOCK_EXCL);
 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 			XFS_QMOPT_RES_REGBLKS);
 	if (error)
 		goto trans_cancel;
-	cancel_flags |= XFS_TRANS_ABORT;
 	if (XFS_IFORK_Q(ip))
 		goto trans_cancel;
 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
@@ -1218,14 +1215,14 @@ xfs_bmap_add_attrfork(
 	error = xfs_bmap_finish(&tp, &flist, &committed);
 	if (error)
 		goto bmap_cancel;
-	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
+	error = xfs_trans_commit(tp);
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;
 
 bmap_cancel:
 	xfs_bmap_cancel(&flist);
 trans_cancel:
-	xfs_trans_cancel(tp, cancel_flags);
+	xfs_trans_cancel(tp);
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;
 }
@@ -3521,7 +3518,8 @@ xfs_bmap_longest_free_extent(
 		}
 	}
 
-	longest = xfs_alloc_longest_free_extent(mp, pag);
+	longest = xfs_alloc_longest_free_extent(mp, pag,
+					xfs_alloc_min_freelist(mp, pag));
 	if (*blen < longest)
 		*blen = longest;
 
@@ -4424,7 +4422,15 @@ xfs_bmapi_convert_unwritten(
 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
 			&bma->cur, mval, bma->firstblock, bma->flist,
 			&tmp_logflags);
-	bma->logflags |= tmp_logflags;
+	/*
+	 * Log the inode core unconditionally in the unwritten extent conversion
+	 * path because the conversion might not have done so (e.g., if the
+	 * extent count hasn't changed). We need to make sure the inode is dirty
+	 * in the transaction for the sake of fsync(), even if nothing has
+	 * changed, because fsync() will not force the log for this transaction
+	 * unless it sees the inode pinned.
+	 */
+	bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
 	if (error)
 		return error;
 
@@ -5918,7 +5924,7 @@ xfs_bmap_split_extent(
 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
 	if (error) {
-		xfs_trans_cancel(tp, 0);
+		xfs_trans_cancel(tp);
 		return error;
 	}
 
@@ -5936,10 +5942,9 @@ xfs_bmap_split_extent(
 	if (error)
 		goto out;
 
-	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
-
+	return xfs_trans_commit(tp);
 
 out:
-	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
+	xfs_trans_cancel(tp);
 	return error;
 }
diff --git a/fs/xfs/libxfs/xfs_format.h b/fs/xfs/libxfs/xfs_format.h
index 4daaa662337b..a0ae572051de 100644
--- a/fs/xfs/libxfs/xfs_format.h
+++ b/fs/xfs/libxfs/xfs_format.h
@@ -170,7 +170,7 @@ typedef struct xfs_sb {
 	__uint32_t	sb_features_log_incompat;
 
 	__uint32_t	sb_crc;		/* superblock crc */
-	__uint32_t	sb_pad;
+	xfs_extlen_t	sb_spino_align;	/* sparse inode chunk alignment */
 
 	xfs_ino_t	sb_pquotino;	/* project quota inode */
 	xfs_lsn_t	sb_lsn;		/* last write sequence */
@@ -256,7 +256,7 @@ typedef struct xfs_dsb {
 	__be32		sb_features_log_incompat;
 
 	__le32		sb_crc;		/* superblock crc */
-	__be32		sb_pad;
+	__be32		sb_spino_align;	/* sparse inode chunk alignment */
 
 	__be64		sb_pquotino;	/* project quota inode */
 	__be64		sb_lsn;		/* last write sequence */
@@ -457,8 +457,10 @@ xfs_sb_has_ro_compat_feature(
 }
 
 #define XFS_SB_FEAT_INCOMPAT_FTYPE	(1 << 0)	/* filetype in dirent */
+#define XFS_SB_FEAT_INCOMPAT_SPINODES	(1 << 1)	/* sparse inode chunks */
 #define XFS_SB_FEAT_INCOMPAT_ALL \
-		(XFS_SB_FEAT_INCOMPAT_FTYPE)
+		(XFS_SB_FEAT_INCOMPAT_FTYPE|	\
+		 XFS_SB_FEAT_INCOMPAT_SPINODES)
 
 #define XFS_SB_FEAT_INCOMPAT_UNKNOWN	~XFS_SB_FEAT_INCOMPAT_ALL
 static inline bool
@@ -506,6 +508,12 @@ static inline int xfs_sb_version_hasfinobt(xfs_sb_t *sbp)
 		(sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT);
 }
 
+static inline bool xfs_sb_version_hassparseinodes(struct xfs_sb *sbp)
+{
+	return XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
+		xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_SPINODES);
+}
+
 /*
  * end of superblock version macros
  */
@@ -758,19 +766,6 @@ typedef struct xfs_agfl {
 
 #define XFS_AGFL_CRC_OFF	offsetof(struct xfs_agfl, agfl_crc)
 
-
-#define	XFS_AG_MAXLEVELS(mp)		((mp)->m_ag_maxlevels)
-#define	XFS_MIN_FREELIST_RAW(bl,cl,mp)	\
-	(MIN(bl + 1, XFS_AG_MAXLEVELS(mp)) + MIN(cl + 1, XFS_AG_MAXLEVELS(mp)))
-#define	XFS_MIN_FREELIST(a,mp)		\
-	(XFS_MIN_FREELIST_RAW(		\
-		be32_to_cpu((a)->agf_levels[XFS_BTNUM_BNOi]), \
-		be32_to_cpu((a)->agf_levels[XFS_BTNUM_CNTi]), mp))
-#define	XFS_MIN_FREELIST_PAG(pag,mp)	\
-	(XFS_MIN_FREELIST_RAW(		\
-		(unsigned int)(pag)->pagf_levels[XFS_BTNUM_BNOi], \
-		(unsigned int)(pag)->pagf_levels[XFS_BTNUM_CNTi], mp))
-
 #define XFS_AGB_TO_FSB(mp,agno,agbno)	\
 	(((xfs_fsblock_t)(agno) << (mp)->m_sb.sb_agblklog) | (agbno))
 #define	XFS_FSB_TO_AGNO(mp,fsbno)	\
@@ -1216,26 +1211,54 @@ typedef	__uint64_t	xfs_inofree_t;
 #define	XFS_INOBT_ALL_FREE		((xfs_inofree_t)-1)
 #define	XFS_INOBT_MASK(i)		((xfs_inofree_t)1 << (i))
 
+#define XFS_INOBT_HOLEMASK_FULL		0	/* holemask for full chunk */
+#define XFS_INOBT_HOLEMASK_BITS		(NBBY * sizeof(__uint16_t))
+#define XFS_INODES_PER_HOLEMASK_BIT	\
+	(XFS_INODES_PER_CHUNK / (NBBY * sizeof(__uint16_t)))
+
 static inline xfs_inofree_t xfs_inobt_maskn(int i, int n)
 {
 	return ((n >= XFS_INODES_PER_CHUNK ? 0 : XFS_INOBT_MASK(n)) - 1) << i;
 }
 
 /*
- * Data record structure
+ * The on-disk inode record structure has two formats. The original "full"
+ * format uses a 4-byte freecount. The "sparse" format uses a 1-byte freecount
+ * and replaces the 3 high-order freecount bytes wth the holemask and inode
+ * count.
+ *
+ * The holemask of the sparse record format allows an inode chunk to have holes
+ * that refer to blocks not owned by the inode record. This facilitates inode
+ * allocation in the event of severe free space fragmentation.
  */
 typedef struct xfs_inobt_rec {
 	__be32		ir_startino;	/* starting inode number */
-	__be32		ir_freecount;	/* count of free inodes (set bits) */
+	union {
+		struct {
+			__be32	ir_freecount;	/* count of free inodes */
+		} f;
+		struct {
+			__be16	ir_holemask;/* hole mask for sparse chunks */
+			__u8	ir_count;	/* total inode count */
+			__u8	ir_freecount;	/* count of free inodes */
+		} sp;
+	} ir_u;
 	__be64		ir_free;	/* free inode mask */
 } xfs_inobt_rec_t;
 
 typedef struct xfs_inobt_rec_incore {
 	xfs_agino_t	ir_startino;	/* starting inode number */
-	__int32_t	ir_freecount;	/* count of free inodes (set bits) */
+	__uint16_t	ir_holemask;	/* hole mask for sparse chunks */
+	__uint8_t	ir_count;	/* total inode count */
+	__uint8_t	ir_freecount;	/* count of free inodes (set bits) */
 	xfs_inofree_t	ir_free;	/* free inode mask */
 } xfs_inobt_rec_incore_t;
 
+static inline bool xfs_inobt_issparse(uint16_t holemask)
+{
+	/* non-zero holemask represents a sparse rec. */
+	return holemask;
+}
 
 /*
  * Key structure
@@ -1453,8 +1476,8 @@ struct xfs_acl {
 		sizeof(struct xfs_acl_entry) * XFS_ACL_MAX_ENTRIES((mp)))
 
 /* On-disk XFS extended attribute names */
-#define SGI_ACL_FILE		(unsigned char *)"SGI_ACL_FILE"
-#define SGI_ACL_DEFAULT		(unsigned char *)"SGI_ACL_DEFAULT"
+#define SGI_ACL_FILE		"SGI_ACL_FILE"
+#define SGI_ACL_DEFAULT		"SGI_ACL_DEFAULT"
 #define SGI_ACL_FILE_SIZE	(sizeof(SGI_ACL_FILE)-1)
 #define SGI_ACL_DEFAULT_SIZE	(sizeof(SGI_ACL_DEFAULT)-1)
 
diff --git a/fs/xfs/libxfs/xfs_fs.h b/fs/xfs/libxfs/xfs_fs.h
index 18dc721ca19f..89689c6a43e2 100644
--- a/fs/xfs/libxfs/xfs_fs.h
+++ b/fs/xfs/libxfs/xfs_fs.h
@@ -239,6 +239,7 @@ typedef struct xfs_fsop_resblks {
 #define XFS_FSOP_GEOM_FLAGS_V5SB	0x8000	/* version 5 superblock */
 #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	*/
 
 /*
  * Minimum and maximum sizes need for growth checks.
diff --git a/fs/xfs/libxfs/xfs_ialloc.c b/fs/xfs/libxfs/xfs_ialloc.c
index 1c9e75521250..66efc702452a 100644
--- a/fs/xfs/libxfs/xfs_ialloc.c
+++ b/fs/xfs/libxfs/xfs_ialloc.c
@@ -65,6 +65,8 @@ xfs_inobt_lookup(
 	int			*stat)	/* success/failure */
 {
 	cur->bc_rec.i.ir_startino = ino;
+	cur->bc_rec.i.ir_holemask = 0;
+	cur->bc_rec.i.ir_count = 0;
 	cur->bc_rec.i.ir_freecount = 0;
 	cur->bc_rec.i.ir_free = 0;
 	return xfs_btree_lookup(cur, dir, stat);
@@ -82,7 +84,14 @@ xfs_inobt_update(
 	union xfs_btree_rec	rec;
 
 	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
-	rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
+	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
+		rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
+		rec.inobt.ir_u.sp.ir_count = irec->ir_count;
+		rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
+	} else {
+		/* ir_holemask/ir_count not supported on-disk */
+		rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
+	}
 	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
 	return xfs_btree_update(cur, &rec);
 }
@@ -100,12 +109,27 @@ xfs_inobt_get_rec(
 	int			error;
 
 	error = xfs_btree_get_rec(cur, &rec, stat);
-	if (!error && *stat == 1) {
-		irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
-		irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
-		irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
+	if (error || *stat == 0)
+		return error;
+
+	irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
+	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
+		irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
+		irec->ir_count = rec->inobt.ir_u.sp.ir_count;
+		irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
+	} else {
+		/*
+		 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
+		 * values for full inode chunks.
+		 */
+		irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
+		irec->ir_count = XFS_INODES_PER_CHUNK;
+		irec->ir_freecount =
+				be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
 	}
-	return error;
+	irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
+
+	return 0;
 }
 
 /*
@@ -114,10 +138,14 @@ xfs_inobt_get_rec(
 STATIC int
 xfs_inobt_insert_rec(
 	struct xfs_btree_cur	*cur,
+	__uint16_t		holemask,
+	__uint8_t		count,
 	__int32_t		freecount,
 	xfs_inofree_t		free,
 	int			*stat)
 {
+	cur->bc_rec.i.ir_holemask = holemask;
+	cur->bc_rec.i.ir_count = count;
 	cur->bc_rec.i.ir_freecount = freecount;
 	cur->bc_rec.i.ir_free = free;
 	return xfs_btree_insert(cur, stat);
@@ -154,7 +182,9 @@ xfs_inobt_insert(
 		}
 		ASSERT(i == 0);
 
-		error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
+		error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
+					     XFS_INODES_PER_CHUNK,
+					     XFS_INODES_PER_CHUNK,
 					     XFS_INOBT_ALL_FREE, &i);
 		if (error) {
 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
@@ -220,6 +250,7 @@ xfs_ialloc_inode_init(
 	struct xfs_mount	*mp,
 	struct xfs_trans	*tp,
 	struct list_head	*buffer_list,
+	int			icount,
 	xfs_agnumber_t		agno,
 	xfs_agblock_t		agbno,
 	xfs_agblock_t		length,
@@ -275,7 +306,7 @@ xfs_ialloc_inode_init(
 		 * they track in the AIL as if they were physically logged.
 		 */
 		if (tp)
-			xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
+			xfs_icreate_log(tp, agno, agbno, icount,
 					mp->m_sb.sb_inodesize, length, gen);
 	} else
 		version = 2;
@@ -347,6 +378,214 @@ xfs_ialloc_inode_init(
 }
 
 /*
+ * Align startino and allocmask for a recently allocated sparse chunk such that
+ * they are fit for insertion (or merge) into the on-disk inode btrees.
+ *
+ * Background:
+ *
+ * When enabled, sparse inode support increases the inode alignment from cluster
+ * size to inode chunk size. This means that the minimum range between two
+ * non-adjacent inode records in the inobt is large enough for a full inode
+ * record. This allows for cluster sized, cluster aligned block allocation
+ * without need to worry about whether the resulting inode record overlaps with
+ * another record in the tree. Without this basic rule, we would have to deal
+ * with the consequences of overlap by potentially undoing recent allocations in
+ * the inode allocation codepath.
+ *
+ * Because of this alignment rule (which is enforced on mount), there are two
+ * inobt possibilities for newly allocated sparse chunks. One is that the
+ * aligned inode record for the chunk covers a range of inodes not already
+ * covered in the inobt (i.e., it is safe to insert a new sparse record). The
+ * other is that a record already exists at the aligned startino that considers
+ * the newly allocated range as sparse. In the latter case, record content is
+ * merged in hope that sparse inode chunks fill to full chunks over time.
+ */
+STATIC void
+xfs_align_sparse_ino(
+	struct xfs_mount		*mp,
+	xfs_agino_t			*startino,
+	uint16_t			*allocmask)
+{
+	xfs_agblock_t			agbno;
+	xfs_agblock_t			mod;
+	int				offset;
+
+	agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
+	mod = agbno % mp->m_sb.sb_inoalignmt;
+	if (!mod)
+		return;
+
+	/* calculate the inode offset and align startino */
+	offset = mod << mp->m_sb.sb_inopblog;
+	*startino -= offset;
+
+	/*
+	 * Since startino has been aligned down, left shift allocmask such that
+	 * it continues to represent the same physical inodes relative to the
+	 * new startino.
+	 */
+	*allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
+}
+
+/*
+ * Determine whether the source inode record can merge into the target. Both
+ * records must be sparse, the inode ranges must match and there must be no
+ * allocation overlap between the records.
+ */
+STATIC bool
+__xfs_inobt_can_merge(
+	struct xfs_inobt_rec_incore	*trec,	/* tgt record */
+	struct xfs_inobt_rec_incore	*srec)	/* src record */
+{
+	uint64_t			talloc;
+	uint64_t			salloc;
+
+	/* records must cover the same inode range */
+	if (trec->ir_startino != srec->ir_startino)
+		return false;
+
+	/* both records must be sparse */
+	if (!xfs_inobt_issparse(trec->ir_holemask) ||
+	    !xfs_inobt_issparse(srec->ir_holemask))
+		return false;
+
+	/* both records must track some inodes */
+	if (!trec->ir_count || !srec->ir_count)
+		return false;
+
+	/* can't exceed capacity of a full record */
+	if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
+		return false;
+
+	/* verify there is no allocation overlap */
+	talloc = xfs_inobt_irec_to_allocmask(trec);
+	salloc = xfs_inobt_irec_to_allocmask(srec);
+	if (talloc & salloc)
+		return false;
+
+	return true;
+}
+
+/*
+ * Merge the source inode record into the target. The caller must call
+ * __xfs_inobt_can_merge() to ensure the merge is valid.
+ */
+STATIC void
+__xfs_inobt_rec_merge(
+	struct xfs_inobt_rec_incore	*trec,	/* target */
+	struct xfs_inobt_rec_incore	*srec)	/* src */
+{
+	ASSERT(trec->ir_startino == srec->ir_startino);
+
+	/* combine the counts */
+	trec->ir_count += srec->ir_count;
+	trec->ir_freecount += srec->ir_freecount;
+
+	/*
+	 * Merge the holemask and free mask. For both fields, 0 bits refer to
+	 * allocated inodes. We combine the allocated ranges with bitwise AND.
+	 */
+	trec->ir_holemask &= srec->ir_holemask;
+	trec->ir_free &= srec->ir_free;
+}
+
+/*
+ * Insert a new sparse inode chunk into the associated inode btree. The inode
+ * record for the sparse chunk is pre-aligned to a startino that should match
+ * any pre-existing sparse inode record in the tree. This allows sparse chunks
+ * to fill over time.
+ *
+ * This function supports two modes of handling preexisting records depending on
+ * the merge flag. If merge is true, the provided record is merged with the
+ * existing record and updated in place. The merged record is returned in nrec.
+ * If merge is false, an existing record is replaced with the provided record.
+ * If no preexisting record exists, the provided record is always inserted.
+ *
+ * It is considered corruption if a merge is requested and not possible. Given
+ * the sparse inode alignment constraints, this should never happen.
+ */
+STATIC int
+xfs_inobt_insert_sprec(
+	struct xfs_mount		*mp,
+	struct xfs_trans		*tp,
+	struct xfs_buf			*agbp,
+	int				btnum,
+	struct xfs_inobt_rec_incore	*nrec,	/* in/out: new/merged rec. */
+	bool				merge)	/* merge or replace */
+{
+	struct xfs_btree_cur		*cur;
+	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
+	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
+	int				error;
+	int				i;
+	struct xfs_inobt_rec_incore	rec;
+
+	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
+
+	/* the new record is pre-aligned so we know where to look */
+	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
+	if (error)
+		goto error;
+	/* if nothing there, insert a new record and return */
+	if (i == 0) {
+		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
+					     nrec->ir_count, nrec->ir_freecount,
+					     nrec->ir_free, &i);
+		if (error)
+			goto error;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
+
+		goto out;
+	}
+
+	/*
+	 * A record exists at this startino. Merge or replace the record
+	 * depending on what we've been asked to do.
+	 */
+	if (merge) {
+		error = xfs_inobt_get_rec(cur, &rec, &i);
+		if (error)
+			goto error;
+		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
+		XFS_WANT_CORRUPTED_GOTO(mp,
+					rec.ir_startino == nrec->ir_startino,
+					error);
+
+		/*
+		 * This should never fail. If we have coexisting records that
+		 * cannot merge, something is seriously wrong.
+		 */
+		XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
+					error);
+
+		trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
+					 rec.ir_holemask, nrec->ir_startino,
+					 nrec->ir_holemask);
+
+		/* merge to nrec to output the updated record */
+		__xfs_inobt_rec_merge(nrec, &rec);
+
+		trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
+					  nrec->ir_holemask);
+
+		error = xfs_inobt_rec_check_count(mp, nrec);
+		if (error)
+			goto error;
+	}
+
+	error = xfs_inobt_update(cur, nrec);
+	if (error)
+		goto error;
+
+out:
+	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+	return 0;
+error:
+	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
+	return error;
+}
+
+/*
  * Allocate new inodes in the allocation group specified by agbp.
  * Return 0 for success, else error code.
  */
@@ -364,11 +603,22 @@ xfs_ialloc_ag_alloc(
 	xfs_agino_t	newlen;		/* new number of inodes */
 	int		isaligned = 0;	/* inode allocation at stripe unit */
 					/* boundary */
+	uint16_t	allocmask = (uint16_t) -1; /* init. to full chunk */
+	struct xfs_inobt_rec_incore rec;
 	struct xfs_perag *pag;
+	int		do_sparse = 0;
 
 	memset(&args, 0, sizeof(args));
 	args.tp = tp;
 	args.mp = tp->t_mountp;
+	args.fsbno = NULLFSBLOCK;
+
+#ifdef DEBUG
+	/* randomly do sparse inode allocations */
+	if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
+	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
+		do_sparse = prandom_u32() & 1;
+#endif
 
 	/*
 	 * Locking will ensure that we don't have two callers in here
@@ -390,6 +640,8 @@ xfs_ialloc_ag_alloc(
 	agno = be32_to_cpu(agi->agi_seqno);
 	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
 		     args.mp->m_ialloc_blks;
+	if (do_sparse)
+		goto sparse_alloc;
 	if (likely(newino != NULLAGINO &&
 		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
@@ -428,8 +680,7 @@ xfs_ialloc_ag_alloc(
 		 * subsequent requests.
 		 */
 		args.minalignslop = 0;
-	} else
-		args.fsbno = NULLFSBLOCK;
+	}
 
 	if (unlikely(args.fsbno == NULLFSBLOCK)) {
 		/*
@@ -480,6 +731,47 @@ xfs_ialloc_ag_alloc(
 			return error;
 	}
 
+	/*
+	 * Finally, try a sparse allocation if the filesystem supports it and
+	 * the sparse allocation length is smaller than a full chunk.
+	 */
+	if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
+	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
+	    args.fsbno == NULLFSBLOCK) {
+sparse_alloc:
+		args.type = XFS_ALLOCTYPE_NEAR_BNO;
+		args.agbno = be32_to_cpu(agi->agi_root);
+		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
+		args.alignment = args.mp->m_sb.sb_spino_align;
+		args.prod = 1;
+
+		args.minlen = args.mp->m_ialloc_min_blks;
+		args.maxlen = args.minlen;
+
+		/*
+		 * The inode record will be aligned to full chunk size. We must
+		 * prevent sparse allocation from AG boundaries that result in
+		 * invalid inode records, such as records that start at agbno 0
+		 * or extend beyond the AG.
+		 *
+		 * Set min agbno to the first aligned, non-zero agbno and max to
+		 * the last aligned agbno that is at least one full chunk from
+		 * the end of the AG.
+		 */
+		args.min_agbno = args.mp->m_sb.sb_inoalignmt;
+		args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
+					    args.mp->m_sb.sb_inoalignmt) -
+				 args.mp->m_ialloc_blks;
+
+		error = xfs_alloc_vextent(&args);
+		if (error)
+			return error;
+
+		newlen = args.len << args.mp->m_sb.sb_inopblog;
+		ASSERT(newlen <= XFS_INODES_PER_CHUNK);
+		allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
+	}
+
 	if (args.fsbno == NULLFSBLOCK) {
 		*alloc = 0;
 		return 0;
@@ -495,8 +787,8 @@ xfs_ialloc_ag_alloc(
 	 * rather than a linear progression to prevent the next generation
 	 * number from being easily guessable.
 	 */
-	error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
-			args.len, prandom_u32());
+	error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
+			args.agbno, args.len, prandom_u32());
 
 	if (error)
 		return error;
@@ -504,6 +796,73 @@ xfs_ialloc_ag_alloc(
 	 * Convert the results.
 	 */
 	newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
+
+	if (xfs_inobt_issparse(~allocmask)) {
+		/*
+		 * We've allocated a sparse chunk. Align the startino and mask.
+		 */
+		xfs_align_sparse_ino(args.mp, &newino, &allocmask);
+
+		rec.ir_startino = newino;
+		rec.ir_holemask = ~allocmask;
+		rec.ir_count = newlen;
+		rec.ir_freecount = newlen;
+		rec.ir_free = XFS_INOBT_ALL_FREE;
+
+		/*
+		 * Insert the sparse record into the inobt and allow for a merge
+		 * if necessary. If a merge does occur, rec is updated to the
+		 * merged record.
+		 */
+		error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
+					       &rec, true);
+		if (error == -EFSCORRUPTED) {
+			xfs_alert(args.mp,
+	"invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
+				  XFS_AGINO_TO_INO(args.mp, agno,
+						   rec.ir_startino),
+				  rec.ir_holemask, rec.ir_count);
+			xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
+		}
+		if (error)
+			return error;
+
+		/*
+		 * We can't merge the part we've just allocated as for the inobt
+		 * due to finobt semantics. The original record may or may not
+		 * exist independent of whether physical inodes exist in this
+		 * sparse chunk.
+		 *
+		 * We must update the finobt record based on the inobt record.
+		 * rec contains the fully merged and up to date inobt record
+		 * from the previous call. Set merge false to replace any
+		 * existing record with this one.
+		 */
+		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
+			error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
+						       XFS_BTNUM_FINO, &rec,
+						       false);
+			if (error)
+				return error;
+		}
+	} else {
+		/* full chunk - insert new records to both btrees */
+		error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
+					 XFS_BTNUM_INO);
+		if (error)
+			return error;
+
+		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
+			error = xfs_inobt_insert(args.mp, tp, agbp, newino,
+						 newlen, XFS_BTNUM_FINO);
+			if (error)
+				return error;
+		}
+	}
+
+	/*
+	 * Update AGI counts and newino.
+	 */
 	be32_add_cpu(&agi->agi_count, newlen);
 	be32_add_cpu(&agi->agi_freecount, newlen);
 	pag = xfs_perag_get(args.mp, agno);
@@ -512,20 +871,6 @@ xfs_ialloc_ag_alloc(
 	agi->agi_newino = cpu_to_be32(newino);
 
 	/*
-	 * Insert records describing the new inode chunk into the btrees.
-	 */
-	error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
-				 XFS_BTNUM_INO);
-	if (error)
-		return error;
-
-	if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
-		error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
-					 XFS_BTNUM_FINO);
-		if (error)
-			return error;
-	}
-	/*
 	 * Log allocation group header fields
 	 */
 	xfs_ialloc_log_agi(tp, agbp,
@@ -645,7 +990,7 @@ xfs_ialloc_ag_select(
 		 * if we fail allocation due to alignment issues then it is most
 		 * likely a real ENOSPC condition.
 		 */
-		ineed = mp->m_ialloc_blks;
+		ineed = mp->m_ialloc_min_blks;
 		if (flags && ineed > 1)
 			ineed += xfs_ialloc_cluster_alignment(mp);
 		longest = pag->pagf_longest;
@@ -732,6 +1077,27 @@ xfs_ialloc_get_rec(
 }
 
 /*
+ * Return the offset of the first free inode in the record. If the inode chunk
+ * is sparsely allocated, we convert the record holemask to inode granularity
+ * and mask off the unallocated regions from the inode free mask.
+ */
+STATIC int
+xfs_inobt_first_free_inode(
+	struct xfs_inobt_rec_incore	*rec)
+{
+	xfs_inofree_t			realfree;
+
+	/* if there are no holes, return the first available offset */
+	if (!xfs_inobt_issparse(rec->ir_holemask))
+		return xfs_lowbit64(rec->ir_free);
+
+	realfree = xfs_inobt_irec_to_allocmask(rec);
+	realfree &= rec->ir_free;
+
+	return xfs_lowbit64(realfree);
+}
+
+/*
  * Allocate an inode using the inobt-only algorithm.
  */
 STATIC int
@@ -961,7 +1327,7 @@ newino:
 	}
 
 alloc_inode:
-	offset = xfs_lowbit64(rec.ir_free);
+	offset = xfs_inobt_first_free_inode(&rec);
 	ASSERT(offset >= 0);
 	ASSERT(offset < XFS_INODES_PER_CHUNK);
 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
@@ -1210,7 +1576,7 @@ xfs_dialloc_ag(
 	if (error)
 		goto error_cur;
 
-	offset = xfs_lowbit64(rec.ir_free);
+	offset = xfs_inobt_first_free_inode(&rec);
 	ASSERT(offset >= 0);
 	ASSERT(offset < XFS_INODES_PER_CHUNK);
 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
@@ -1439,6 +1805,83 @@ out_error:
 	return error;
 }
 
+/*
+ * Free the blocks of an inode chunk. We must consider that the inode chunk
+ * might be sparse and only free the regions that are allocated as part of the
+ * chunk.
+ */
+STATIC void
+xfs_difree_inode_chunk(
+	struct xfs_mount		*mp,
+	xfs_agnumber_t			agno,
+	struct xfs_inobt_rec_incore	*rec,
+	struct xfs_bmap_free		*flist)
+{
+	xfs_agblock_t	sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino);
+	int		startidx, endidx;
+	int		nextbit;
+	xfs_agblock_t	agbno;
+	int		contigblk;
+	DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
+
+	if (!xfs_inobt_issparse(rec->ir_holemask)) {
+		/* not sparse, calculate extent info directly */
+		xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
+				  XFS_AGINO_TO_AGBNO(mp, rec->ir_startino)),
+				  mp->m_ialloc_blks, flist, mp);
+		return;
+	}
+
+	/* holemask is only 16-bits (fits in an unsigned long) */
+	ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
+	holemask[0] = rec->ir_holemask;
+
+	/*
+	 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
+	 * holemask and convert the start/end index of each range to an extent.
+	 * We start with the start and end index both pointing at the first 0 in
+	 * the mask.
+	 */
+	startidx = endidx = find_first_zero_bit(holemask,
+						XFS_INOBT_HOLEMASK_BITS);
+	nextbit = startidx + 1;
+	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
+		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
+					     nextbit);
+		/*
+		 * If the next zero bit is contiguous, update the end index of
+		 * the current range and continue.
+		 */
+		if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
+		    nextbit == endidx + 1) {
+			endidx = nextbit;
+			goto next;
+		}
+
+		/*
+		 * nextbit is not contiguous with the current end index. Convert
+		 * the current start/end to an extent and add it to the free
+		 * list.
+		 */
+		agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
+				  mp->m_sb.sb_inopblock;
+		contigblk = ((endidx - startidx + 1) *
+			     XFS_INODES_PER_HOLEMASK_BIT) /
+			    mp->m_sb.sb_inopblock;
+
+		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
+		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
+		xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, agbno), contigblk,
+				  flist, mp);
+
+		/* reset range to current bit and carry on... */
+		startidx = endidx = nextbit;
+
+next:
+		nextbit++;
+	}
+}
+
 STATIC int
 xfs_difree_inobt(
 	struct xfs_mount		*mp,
@@ -1446,8 +1889,7 @@ xfs_difree_inobt(
 	struct xfs_buf			*agbp,
 	xfs_agino_t			agino,
 	struct xfs_bmap_free		*flist,
-	int				*deleted,
-	xfs_ino_t			*first_ino,
+	struct xfs_icluster		*xic,
 	struct xfs_inobt_rec_incore	*orec)
 {
 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
@@ -1501,20 +1943,23 @@ xfs_difree_inobt(
 	rec.ir_freecount++;
 
 	/*
-	 * When an inode cluster is free, it becomes eligible for removal
+	 * When an inode chunk is free, it becomes eligible for removal. Don't
+	 * remove the chunk if the block size is large enough for multiple inode
+	 * chunks (that might not be free).
 	 */
 	if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
-	    (rec.ir_freecount == mp->m_ialloc_inos)) {
-
-		*deleted = 1;
-		*first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
+	    rec.ir_free == XFS_INOBT_ALL_FREE &&
+	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
+		xic->deleted = 1;
+		xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
+		xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
 
 		/*
 		 * Remove the inode cluster from the AGI B+Tree, adjust the
 		 * AGI and Superblock inode counts, and mark the disk space
 		 * to be freed when the transaction is committed.
 		 */
-		ilen = mp->m_ialloc_inos;
+		ilen = rec.ir_freecount;
 		be32_add_cpu(&agi->agi_count, -ilen);
 		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
@@ -1530,11 +1975,9 @@ xfs_difree_inobt(
 			goto error0;
 		}
 
-		xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
-				  XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
-				  mp->m_ialloc_blks, flist, mp);
+		xfs_difree_inode_chunk(mp, agno, &rec, flist);
 	} else {
-		*deleted = 0;
+		xic->deleted = 0;
 
 		error = xfs_inobt_update(cur, &rec);
 		if (error) {
@@ -1599,7 +2042,9 @@ xfs_difree_finobt(
 		 */
 		XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
 
-		error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
+		error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
+					     ibtrec->ir_count,
+					     ibtrec->ir_freecount,
 					     ibtrec->ir_free, &i);
 		if (error)
 			goto error;
@@ -1634,8 +2079,13 @@ xfs_difree_finobt(
 	 * free inode. Hence, if all of the inodes are free and we aren't
 	 * keeping inode chunks permanently on disk, remove the record.
 	 * Otherwise, update the record with the new information.
+	 *
+	 * Note that we currently can't free chunks when the block size is large
+	 * enough for multiple chunks. Leave the finobt record to remain in sync
+	 * with the inobt.
 	 */
-	if (rec.ir_freecount == mp->m_ialloc_inos &&
+	if (rec.ir_free == XFS_INOBT_ALL_FREE &&
+	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
 	    !(mp->m_flags & XFS_MOUNT_IKEEP)) {
 		error = xfs_btree_delete(cur, &i);
 		if (error)
@@ -1671,8 +2121,7 @@ xfs_difree(
 	struct xfs_trans	*tp,		/* transaction pointer */
 	xfs_ino_t		inode,		/* inode to be freed */
 	struct xfs_bmap_free	*flist,		/* extents to free */
-	int			*deleted,/* set if inode cluster was deleted */
-	xfs_ino_t		*first_ino)/* first inode in deleted cluster */
+	struct xfs_icluster	*xic)	/* cluster info if deleted */
 {
 	/* REFERENCED */
 	xfs_agblock_t		agbno;	/* block number containing inode */
@@ -1723,8 +2172,7 @@ xfs_difree(
 	/*
 	 * Fix up the inode allocation btree.
 	 */
-	error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino,
-				 &rec);
+	error = xfs_difree_inobt(mp, tp, agbp, agino, flist, xic, &rec);
 	if (error)
 		goto error0;
 
diff --git a/fs/xfs/libxfs/xfs_ialloc.h b/fs/xfs/libxfs/xfs_ialloc.h
index 100007d56449..6e450df2979b 100644
--- a/fs/xfs/libxfs/xfs_ialloc.h
+++ b/fs/xfs/libxfs/xfs_ialloc.h
@@ -28,6 +28,13 @@ struct xfs_btree_cur;
 /* Move inodes in clusters of this size */
 #define	XFS_INODE_BIG_CLUSTER_SIZE	8192
 
+struct xfs_icluster {
+	bool		deleted;	/* record is deleted */
+	xfs_ino_t	first_ino;	/* first inode number */
+	uint64_t	alloc;		/* inode phys. allocation bitmap for
+					 * sparse chunks */
+};
+
 /* Calculate and return the number of filesystem blocks per inode cluster */
 static inline int
 xfs_icluster_size_fsb(
@@ -44,8 +51,7 @@ xfs_icluster_size_fsb(
 static inline struct xfs_dinode *
 xfs_make_iptr(struct xfs_mount *mp, struct xfs_buf *b, int o)
 {
-	return (struct xfs_dinode *)
-		(xfs_buf_offset(b, o << (mp)->m_sb.sb_inodelog));
+	return xfs_buf_offset(b, o << (mp)->m_sb.sb_inodelog);
 }
 
 /*
@@ -90,8 +96,7 @@ xfs_difree(
 	struct xfs_trans *tp,		/* transaction pointer */
 	xfs_ino_t	inode,		/* inode to be freed */
 	struct xfs_bmap_free *flist,	/* extents to free */
-	int		*deleted,	/* set if inode cluster was deleted */
-	xfs_ino_t	*first_ino);	/* first inode in deleted cluster */
+	struct xfs_icluster *ifree);	/* cluster info if deleted */
 
 /*
  * Return the location of the inode in imap, for mapping it into a buffer.
@@ -156,7 +161,7 @@ int xfs_inobt_get_rec(struct xfs_btree_cur *cur,
  * Inode chunk initialisation routine
  */
 int xfs_ialloc_inode_init(struct xfs_mount *mp, struct xfs_trans *tp,
-			  struct list_head *buffer_list,
+			  struct list_head *buffer_list, int icount,
 			  xfs_agnumber_t agno, xfs_agblock_t agbno,
 			  xfs_agblock_t length, unsigned int gen);
 
diff --git a/fs/xfs/libxfs/xfs_ialloc_btree.c b/fs/xfs/libxfs/xfs_ialloc_btree.c
index 964c465ca69c..674ad8f760be 100644
--- a/fs/xfs/libxfs/xfs_ialloc_btree.c
+++ b/fs/xfs/libxfs/xfs_ialloc_btree.c
@@ -167,7 +167,16 @@ xfs_inobt_init_rec_from_cur(
 	union xfs_btree_rec	*rec)
 {
 	rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
-	rec->inobt.ir_freecount = cpu_to_be32(cur->bc_rec.i.ir_freecount);
+	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
+		rec->inobt.ir_u.sp.ir_holemask =
+					cpu_to_be16(cur->bc_rec.i.ir_holemask);
+		rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
+		rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
+	} else {
+		/* ir_holemask/ir_count not supported on-disk */
+		rec->inobt.ir_u.f.ir_freecount =
+					cpu_to_be32(cur->bc_rec.i.ir_freecount);
+	}
 	rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
 }
 
@@ -418,3 +427,85 @@ xfs_inobt_maxrecs(
 		return blocklen / sizeof(xfs_inobt_rec_t);
 	return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
 }
+
+/*
+ * Convert the inode record holemask to an inode allocation bitmap. The inode
+ * allocation bitmap is inode granularity and specifies whether an inode is
+ * physically allocated on disk (not whether the inode is considered allocated
+ * or free by the fs).
+ *
+ * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
+ */
+uint64_t
+xfs_inobt_irec_to_allocmask(
+	struct xfs_inobt_rec_incore	*rec)
+{
+	uint64_t			bitmap = 0;
+	uint64_t			inodespbit;
+	int				nextbit;
+	uint				allocbitmap;
+
+	/*
+	 * The holemask has 16-bits for a 64 inode record. Therefore each
+	 * holemask bit represents multiple inodes. Create a mask of bits to set
+	 * in the allocmask for each holemask bit.
+	 */
+	inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
+
+	/*
+	 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
+	 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
+	 * anything beyond the 16 holemask bits since this casts to a larger
+	 * type.
+	 */
+	allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
+
+	/*
+	 * allocbitmap is the inverted holemask so every set bit represents
+	 * allocated inodes. To expand from 16-bit holemask granularity to
+	 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
+	 * bitmap for every holemask bit.
+	 */
+	nextbit = xfs_next_bit(&allocbitmap, 1, 0);
+	while (nextbit != -1) {
+		ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
+
+		bitmap |= (inodespbit <<
+			   (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
+
+		nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
+	}
+
+	return bitmap;
+}
+
+#if defined(DEBUG) || defined(XFS_WARN)
+/*
+ * Verify that an in-core inode record has a valid inode count.
+ */
+int
+xfs_inobt_rec_check_count(
+	struct xfs_mount		*mp,
+	struct xfs_inobt_rec_incore	*rec)
+{
+	int				inocount = 0;
+	int				nextbit = 0;
+	uint64_t			allocbmap;
+	int				wordsz;
+
+	wordsz = sizeof(allocbmap) / sizeof(unsigned int);
+	allocbmap = xfs_inobt_irec_to_allocmask(rec);
+
+	nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
+	while (nextbit != -1) {
+		inocount++;
+		nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
+				       nextbit + 1);
+	}
+
+	if (inocount != rec->ir_count)
+		return -EFSCORRUPTED;
+
+	return 0;
+}
+#endif	/* DEBUG */
diff --git a/fs/xfs/libxfs/xfs_ialloc_btree.h b/fs/xfs/libxfs/xfs_ialloc_btree.h
index d7ebea72c2d0..bd88453217ce 100644
--- a/fs/xfs/libxfs/xfs_ialloc_btree.h
+++ b/fs/xfs/libxfs/xfs_ialloc_btree.h
@@ -62,4 +62,14 @@ extern struct xfs_btree_cur *xfs_inobt_init_cursor(struct xfs_mount *,
 		xfs_btnum_t);
 extern int xfs_inobt_maxrecs(struct xfs_mount *, int, int);
 
+/* ir_holemask to inode allocation bitmap conversion */
+uint64_t xfs_inobt_irec_to_allocmask(struct xfs_inobt_rec_incore *);
+
+#if defined(DEBUG) || defined(XFS_WARN)
+int xfs_inobt_rec_check_count(struct xfs_mount *,
+			      struct xfs_inobt_rec_incore *);
+#else
+#define xfs_inobt_rec_check_count(mp, rec)	0
+#endif	/* DEBUG */
+
 #endif	/* __XFS_IALLOC_BTREE_H__ */
diff --git a/fs/xfs/libxfs/xfs_inode_buf.c b/fs/xfs/libxfs/xfs_inode_buf.c
index 002b6b3a1988..6526e7696184 100644
--- a/fs/xfs/libxfs/xfs_inode_buf.c
+++ b/fs/xfs/libxfs/xfs_inode_buf.c
@@ -46,8 +46,7 @@ xfs_inobp_check(
 	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
 
 	for (i = 0; i < j; i++) {
-		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
-					i * mp->m_sb.sb_inodesize);
+		dip = xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize);
 		if (!dip->di_next_unlinked)  {
 			xfs_alert(mp,
 	"Detected bogus zero next_unlinked field in inode %d buffer 0x%llx.",
@@ -86,8 +85,7 @@ xfs_inode_buf_verify(
 		int		di_ok;
 		xfs_dinode_t	*dip;
 
-		dip = (struct xfs_dinode *)xfs_buf_offset(bp,
-					(i << mp->m_sb.sb_inodelog));
+		dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
 		di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
 			    XFS_DINODE_GOOD_VERSION(dip->di_version);
 		if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
@@ -186,7 +184,7 @@ xfs_imap_to_bp(
 	}
 
 	*bpp = bp;
-	*dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset);
+	*dipp = xfs_buf_offset(bp, imap->im_boffset);
 	return 0;
 }
 
diff --git a/fs/xfs/libxfs/xfs_sb.c b/fs/xfs/libxfs/xfs_sb.c
index dc4bfc5d88fc..df9851c46b5c 100644
--- a/fs/xfs/libxfs/xfs_sb.c
+++ b/fs/xfs/libxfs/xfs_sb.c
@@ -174,6 +174,27 @@ xfs_mount_validate_sb(
 			return -EFSCORRUPTED;
 	}
 
+	/*
+	 * Full inode chunks must be aligned to inode chunk size when
+	 * sparse inodes are enabled to support the sparse chunk
+	 * allocation algorithm and prevent overlapping inode records.
+	 */
+	if (xfs_sb_version_hassparseinodes(sbp)) {
+		uint32_t	align;
+
+		xfs_alert(mp,
+	"EXPERIMENTAL sparse inode feature enabled. Use at your own risk!");
+
+		align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
+				>> sbp->sb_blocklog;
+		if (sbp->sb_inoalignmt != align) {
+			xfs_warn(mp,
+"Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
+				 sbp->sb_inoalignmt, align);
+			return -EINVAL;
+		}
+	}
+
 	if (unlikely(
 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
 		xfs_warn(mp,
@@ -374,7 +395,7 @@ __xfs_sb_from_disk(
 				be32_to_cpu(from->sb_features_log_incompat);
 	/* crc is only used on disk, not in memory; just init to 0 here. */
 	to->sb_crc = 0;
-	to->sb_pad = 0;
+	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
 	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
 	to->sb_lsn = be64_to_cpu(from->sb_lsn);
 	/* Convert on-disk flags to in-memory flags? */
@@ -516,7 +537,7 @@ xfs_sb_to_disk(
 				cpu_to_be32(from->sb_features_incompat);
 		to->sb_features_log_incompat =
 				cpu_to_be32(from->sb_features_log_incompat);
-		to->sb_pad = 0;
+		to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
 		to->sb_lsn = cpu_to_be64(from->sb_lsn);
 	}
 }
@@ -689,6 +710,11 @@ xfs_sb_mount_common(
 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
 					sbp->sb_inopblock);
 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
+
+	if (sbp->sb_spino_align)
+		mp->m_ialloc_min_blks = sbp->sb_spino_align;
+	else
+		mp->m_ialloc_min_blks = mp->m_ialloc_blks;
 }
 
 /*
@@ -792,12 +818,12 @@ xfs_sync_sb(
 	tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_CHANGE, KM_SLEEP);
 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 	if (error) {
-		xfs_trans_cancel(tp, 0);
+		xfs_trans_cancel(tp);
 		return error;
 	}
 
 	xfs_log_sb(tp);
 	if (wait)
 		xfs_trans_set_sync(tp);
-	return xfs_trans_commit(tp, 0);
+	return xfs_trans_commit(tp);
 }
diff --git a/fs/xfs/libxfs/xfs_shared.h b/fs/xfs/libxfs/xfs_shared.h
index 8dda4b321343..5be529707903 100644
--- a/fs/xfs/libxfs/xfs_shared.h
+++ b/fs/xfs/libxfs/xfs_shared.h
@@ -182,12 +182,6 @@ int	xfs_log_calc_minimum_size(struct xfs_mount *);
 #define XFS_TRANS_FREEZE_PROT	0x40	/* Transaction has elevated writer
 					   count in superblock */
 /*
- * Values for call flags parameter.
- */
-#define	XFS_TRANS_RELEASE_LOG_RES	0x4
-#define	XFS_TRANS_ABORT			0x8
-
-/*
  * Field values for xfs_trans_mod_sb.
  */
 #define	XFS_TRANS_SB_ICOUNT		0x00000001
diff --git a/fs/xfs/libxfs/xfs_trans_resv.h b/fs/xfs/libxfs/xfs_trans_resv.h
index 2d5bdfce6d8f..797815012c0e 100644
--- a/fs/xfs/libxfs/xfs_trans_resv.h
+++ b/fs/xfs/libxfs/xfs_trans_resv.h
@@ -73,9 +73,9 @@ struct xfs_trans_resv {
  * 2 trees * (2 blocks/level * max depth - 1) * block size
  */
 #define	XFS_ALLOCFREE_LOG_RES(mp,nx) \
-	((nx) * (2 * XFS_FSB_TO_B((mp), 2 * XFS_AG_MAXLEVELS(mp) - 1)))
+	((nx) * (2 * XFS_FSB_TO_B((mp), 2 * (mp)->m_ag_maxlevels - 1)))
 #define	XFS_ALLOCFREE_LOG_COUNT(mp,nx) \
-	((nx) * (2 * (2 * XFS_AG_MAXLEVELS(mp) - 1)))
+	((nx) * (2 * (2 * (mp)->m_ag_maxlevels - 1)))
 
 /*
  * Per-directory log reservation for any directory change.
diff --git a/fs/xfs/libxfs/xfs_trans_space.h b/fs/xfs/libxfs/xfs_trans_space.h
index bf9c4579334d..41e0428d8175 100644
--- a/fs/xfs/libxfs/xfs_trans_space.h
+++ b/fs/xfs/libxfs/xfs_trans_space.h
@@ -67,7 +67,7 @@
 #define	XFS_DIOSTRAT_SPACE_RES(mp, v)	\
 	(XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK) + (v))
 #define	XFS_GROWFS_SPACE_RES(mp)	\
-	(2 * XFS_AG_MAXLEVELS(mp))
+	(2 * (mp)->m_ag_maxlevels)
 #define	XFS_GROWFSRT_SPACE_RES(mp,b)	\
 	((b) + XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK))
 #define	XFS_LINK_SPACE_RES(mp,nl)	\