[XFS] implement generic xfs_btree_rshift

Make the btree right shift code generic. Based on a patch from David
Chinner with lots of changes to follow the original btree implementations
more closely. While this loses some of the generic helper routines for
inserting/moving/removing records it also solves some of the one off bugs
in the original code and makes it easier to verify.

SGI-PV: 985583

SGI-Modid: xfs-linux-melb:xfs-kern:32196a

Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Bill O'Donnell <billodo@sgi.com>
Signed-off-by: David Chinner <david@fromorbit.com>

5 files changed, 331 insertions, 408 deletions

diff --git a/fs/xfs/xfs_alloc_btree.c b/fs/xfs/xfs_alloc_btree.c
index c5c32999b81..31e42891fc9 100644
--- a/fs/xfs/xfs_alloc_btree.c
+++ b/fs/xfs/xfs_alloc_btree.c
@@ -49,7 +49,6 @@ STATIC void xfs_alloc_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_alloc_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC int xfs_alloc_lshift(xfs_btree_cur_t *, int, int *);
 STATIC int xfs_alloc_newroot(xfs_btree_cur_t *, int *);
-STATIC int xfs_alloc_rshift(xfs_btree_cur_t *, int, int *);
 STATIC int xfs_alloc_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
 		xfs_alloc_key_t *, xfs_btree_cur_t **, int *);
 
@@ -391,7 +390,7 @@ xfs_alloc_delrec(
 		 */
 		if (be16_to_cpu(left->bb_numrecs) - 1 >=
 		     XFS_ALLOC_BLOCK_MINRECS(level, cur)) {
-			if ((error = xfs_alloc_rshift(tcur, level, &i)))
+			if ((error = xfs_btree_rshift(tcur, level, &i)))
 				goto error0;
 			if (i) {
 				ASSERT(be16_to_cpu(block->bb_numrecs) >=
@@ -683,7 +682,7 @@ xfs_alloc_insrec(
 		/*
 		 * First, try shifting an entry to the right neighbor.
 		 */
-		if ((error = xfs_alloc_rshift(cur, level, &i)))
+		if ((error = xfs_btree_rshift(cur, level, &i)))
 			return error;
 		if (i) {
 			/* nothing */
@@ -1233,137 +1232,6 @@ xfs_alloc_newroot(
 }
 
 /*
- * Move 1 record right from cur/level if possible.
- * Update cur to reflect the new path.
- */
-STATIC int				/* error */
-xfs_alloc_rshift(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	int			level,	/* level to shift record on */
-	int			*stat)	/* success/failure */
-{
-	int			error;	/* error return value */
-	int			i;	/* loop index */
-	xfs_alloc_key_t		key;	/* key value for leaf level upward */
-	xfs_buf_t		*lbp;	/* buffer for left (current) block */
-	xfs_alloc_block_t	*left;	/* left (current) btree block */
-	xfs_buf_t		*rbp;	/* buffer for right neighbor block */
-	xfs_alloc_block_t	*right;	/* right neighbor btree block */
-	xfs_alloc_key_t		*rkp;	/* key pointer for right block */
-	xfs_btree_cur_t		*tcur;	/* temporary cursor */
-
-	/*
-	 * Set up variables for this block as "left".
-	 */
-	lbp = cur->bc_bufs[level];
-	left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
-#ifdef DEBUG
-	if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-		return error;
-#endif
-	/*
-	 * If we've got no right sibling then we can't shift an entry right.
-	 */
-	if (be32_to_cpu(left->bb_rightsib) == NULLAGBLOCK) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * If the cursor entry is the one that would be moved, don't
-	 * do it... it's too complicated.
-	 */
-	if (cur->bc_ptrs[level] >= be16_to_cpu(left->bb_numrecs)) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * Set up the right neighbor as "right".
-	 */
-	if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
-			cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib),
-			0, &rbp, XFS_ALLOC_BTREE_REF)))
-		return error;
-	right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
-	if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
-		return error;
-	/*
-	 * If it's full, it can't take another entry.
-	 */
-	if (be16_to_cpu(right->bb_numrecs) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * Make a hole at the start of the right neighbor block, then
-	 * copy the last left block entry to the hole.
-	 */
-	if (level > 0) {
-		xfs_alloc_key_t	*lkp;	/* key pointer for left block */
-		xfs_alloc_ptr_t	*lpp;	/* address pointer for left block */
-		xfs_alloc_ptr_t	*rpp;	/* address pointer for right block */
-
-		lkp = XFS_ALLOC_KEY_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		lpp = XFS_ALLOC_PTR_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
-		rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
-#ifdef DEBUG
-		for (i = be16_to_cpu(right->bb_numrecs) - 1; i >= 0; i--) {
-			if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
-				return error;
-		}
-#endif
-		memmove(rkp + 1, rkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-		memmove(rpp + 1, rpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-#ifdef DEBUG
-		if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(*lpp), level)))
-			return error;
-#endif
-		*rkp = *lkp;
-		*rpp = *lpp;
-		xfs_alloc_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		xfs_alloc_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		xfs_btree_check_key(cur->bc_btnum, rkp, rkp + 1);
-	} else {
-		xfs_alloc_rec_t	*lrp;	/* record pointer for left block */
-		xfs_alloc_rec_t	*rrp;	/* record pointer for right block */
-
-		lrp = XFS_ALLOC_REC_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
-		memmove(rrp + 1, rrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-		*rrp = *lrp;
-		xfs_alloc_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		key.ar_startblock = rrp->ar_startblock;
-		key.ar_blockcount = rrp->ar_blockcount;
-		rkp = &key;
-		xfs_btree_check_rec(cur->bc_btnum, rrp, rrp + 1);
-	}
-	/*
-	 * Decrement and log left's numrecs, bump and log right's numrecs.
-	 */
-	be16_add_cpu(&left->bb_numrecs, -1);
-	xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS);
-	be16_add_cpu(&right->bb_numrecs, 1);
-	xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS);
-	/*
-	 * Using a temporary cursor, update the parent key values of the
-	 * block on the right.
-	 */
-	if ((error = xfs_btree_dup_cursor(cur, &tcur)))
-		return error;
-	i = xfs_btree_lastrec(tcur, level);
-	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-	if ((error = xfs_btree_increment(tcur, level, &i)) ||
-	    (error = xfs_btree_updkey(tcur, (union xfs_btree_key *)rkp, level + 1)))
-		goto error0;
-	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
-	*stat = 1;
-	return 0;
-error0:
-	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
-	return error;
-}
-
-/*
  * Split cur/level block in half.
  * Return new block number and its first record (to be inserted into parent).
  */
diff --git a/fs/xfs/xfs_bmap_btree.c b/fs/xfs/xfs_bmap_btree.c
index 99200a9898f..9d18fa8aa1d 100644
--- a/fs/xfs/xfs_bmap_btree.c
+++ b/fs/xfs/xfs_bmap_btree.c
@@ -53,7 +53,6 @@ STATIC int xfs_bmbt_killroot(xfs_btree_cur_t *);
 STATIC void xfs_bmbt_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_bmbt_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC int xfs_bmbt_lshift(xfs_btree_cur_t *, int, int *);
-STATIC int xfs_bmbt_rshift(xfs_btree_cur_t *, int, int *);
 STATIC int xfs_bmbt_split(xfs_btree_cur_t *, int, xfs_fsblock_t *,
 		__uint64_t *, xfs_btree_cur_t **, int *);
 
@@ -327,7 +326,7 @@ xfs_bmbt_delrec(
 		bno = be64_to_cpu(left->bb_rightsib);
 		if (be16_to_cpu(left->bb_numrecs) - 1 >=
 		    XFS_BMAP_BLOCK_IMINRECS(level, cur)) {
-			if ((error = xfs_bmbt_rshift(tcur, level, &i))) {
+			if ((error = xfs_btree_rshift(tcur, level, &i))) {
 				XFS_BMBT_TRACE_CURSOR(cur, ERROR);
 				goto error0;
 			}
@@ -538,7 +537,7 @@ xfs_bmbt_insrec(
 				logflags);
 			block = xfs_bmbt_get_block(cur, level, &bp);
 		} else {
-			if ((error = xfs_bmbt_rshift(cur, level, &i))) {
+			if ((error = xfs_btree_rshift(cur, level, &i))) {
 				XFS_BMBT_TRACE_CURSOR(cur, ERROR);
 				return error;
 			}
@@ -945,143 +944,6 @@ xfs_bmbt_lshift(
 }
 
 /*
- * Move 1 record right from cur/level if possible.
- * Update cur to reflect the new path.
- */
-STATIC int					/* error */
-xfs_bmbt_rshift(
-	xfs_btree_cur_t		*cur,
-	int			level,
-	int			*stat)		/* success/failure */
-{
-	int			error;		/* error return value */
-	int			i;		/* loop counter */
-	xfs_bmbt_key_t		key;		/* bmap btree key */
-	xfs_buf_t		*lbp;		/* left buffer pointer */
-	xfs_bmbt_block_t	*left;		/* left btree block */
-	xfs_bmbt_key_t		*lkp;		/* left btree key */
-	xfs_bmbt_ptr_t		*lpp;		/* left address pointer */
-	xfs_bmbt_rec_t		*lrp;		/* left record pointer */
-	xfs_mount_t		*mp;		/* file system mount point */
-	xfs_buf_t		*rbp;		/* right buffer pointer */
-	xfs_bmbt_block_t	*right;		/* right btree block */
-	xfs_bmbt_key_t		*rkp;		/* right btree key */
-	xfs_bmbt_ptr_t		*rpp;		/* right address pointer */
-	xfs_bmbt_rec_t		*rrp=NULL;	/* right record pointer */
-	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
-
-	XFS_BMBT_TRACE_CURSOR(cur, ENTRY);
-	XFS_BMBT_TRACE_ARGI(cur, level);
-	if (level == cur->bc_nlevels - 1) {
-		XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-		*stat = 0;
-		return 0;
-	}
-	lbp = cur->bc_bufs[level];
-	left = XFS_BUF_TO_BMBT_BLOCK(lbp);
-#ifdef DEBUG
-	if ((error = xfs_btree_check_lblock(cur, left, level, lbp))) {
-		XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-		return error;
-	}
-#endif
-	if (be64_to_cpu(left->bb_rightsib) == NULLDFSBNO) {
-		XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-		*stat = 0;
-		return 0;
-	}
-	if (cur->bc_ptrs[level] >= be16_to_cpu(left->bb_numrecs)) {
-		XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-		*stat = 0;
-		return 0;
-	}
-	mp = cur->bc_mp;
-	if ((error = xfs_btree_read_bufl(mp, cur->bc_tp, be64_to_cpu(left->bb_rightsib), 0,
-			&rbp, XFS_BMAP_BTREE_REF))) {
-		XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-		return error;
-	}
-	right = XFS_BUF_TO_BMBT_BLOCK(rbp);
-	if ((error = xfs_btree_check_lblock(cur, right, level, rbp))) {
-		XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-		return error;
-	}
-	if (be16_to_cpu(right->bb_numrecs) == XFS_BMAP_BLOCK_IMAXRECS(level, cur)) {
-		XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-		*stat = 0;
-		return 0;
-	}
-	if (level > 0) {
-		lkp = XFS_BMAP_KEY_IADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		lpp = XFS_BMAP_PTR_IADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rkp = XFS_BMAP_KEY_IADDR(right, 1, cur);
-		rpp = XFS_BMAP_PTR_IADDR(right, 1, cur);
-#ifdef DEBUG
-		for (i = be16_to_cpu(right->bb_numrecs) - 1; i >= 0; i--) {
-			if ((error = xfs_btree_check_lptr_disk(cur, rpp[i], level))) {
-				XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-				return error;
-			}
-		}
-#endif
-		memmove(rkp + 1, rkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-		memmove(rpp + 1, rpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-#ifdef DEBUG
-		if ((error = xfs_btree_check_lptr_disk(cur, *lpp, level))) {
-			XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-			return error;
-		}
-#endif
-		*rkp = *lkp;
-		*rpp = *lpp;
-		xfs_bmbt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		xfs_bmbt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-	} else {
-		lrp = XFS_BMAP_REC_IADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rrp = XFS_BMAP_REC_IADDR(right, 1, cur);
-		memmove(rrp + 1, rrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-		*rrp = *lrp;
-		xfs_bmbt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		key.br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(rrp));
-		rkp = &key;
-	}
-	be16_add_cpu(&left->bb_numrecs, -1);
-	xfs_bmbt_log_block(cur, lbp, XFS_BB_NUMRECS);
-	be16_add_cpu(&right->bb_numrecs, 1);
-#ifdef DEBUG
-	if (level > 0)
-		xfs_btree_check_key(XFS_BTNUM_BMAP, rkp, rkp + 1);
-	else
-		xfs_btree_check_rec(XFS_BTNUM_BMAP, rrp, rrp + 1);
-#endif
-	xfs_bmbt_log_block(cur, rbp, XFS_BB_NUMRECS);
-	if ((error = xfs_btree_dup_cursor(cur, &tcur))) {
-		XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-		return error;
-	}
-	i = xfs_btree_lastrec(tcur, level);
-	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-	if ((error = xfs_btree_increment(tcur, level, &i))) {
-		XFS_BMBT_TRACE_CURSOR(tcur, ERROR);
-		goto error1;
-	}
-	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-	if ((error = xfs_btree_updkey(tcur, (union xfs_btree_key *)rkp, level + 1))) {
-		XFS_BMBT_TRACE_CURSOR(tcur, ERROR);
-		goto error1;
-	}
-	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
-	XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-	*stat = 1;
-	return 0;
-error0:
-	XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-error1:
-	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
-	return error;
-}
-
-/*
  * Determine the extent state.
  */
 /* ARGSUSED */
diff --git a/fs/xfs/xfs_btree.c b/fs/xfs/xfs_btree.c
index 205272f282d..e1a21378184 100644
--- a/fs/xfs/xfs_btree.c
+++ b/fs/xfs/xfs_btree.c
@@ -1118,6 +1118,77 @@ xfs_btree_copy_recs(
 }
 
 /*
+ * Copy block pointers from one btree block to another.
+ */
+STATIC void
+xfs_btree_copy_ptrs(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*dst_ptr,
+	union xfs_btree_ptr	*src_ptr,
+	int			numptrs)
+{
+	ASSERT(numptrs >= 0);
+	memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
+}
+
+/*
+ * Shift keys one index left/right inside a single btree block.
+ */
+STATIC void
+xfs_btree_shift_keys(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_key	*key,
+	int			dir,
+	int			numkeys)
+{
+	char			*dst_key;
+
+	ASSERT(numkeys >= 0);
+	ASSERT(dir == 1 || dir == -1);
+
+	dst_key = (char *)key + (dir * cur->bc_ops->key_len);
+	memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
+}
+
+/*
+ * Shift records one index left/right inside a single btree block.
+ */
+STATIC void
+xfs_btree_shift_recs(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_rec	*rec,
+	int			dir,
+	int			numrecs)
+{
+	char			*dst_rec;
+
+	ASSERT(numrecs >= 0);
+	ASSERT(dir == 1 || dir == -1);
+
+	dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
+	memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
+}
+
+/*
+ * Shift block pointers one index left/right inside a single btree block.
+ */
+STATIC void
+xfs_btree_shift_ptrs(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*ptr,
+	int			dir,
+	int			numptrs)
+{
+	char			*dst_ptr;
+
+	ASSERT(numptrs >= 0);
+	ASSERT(dir == 1 || dir == -1);
+
+	dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
+	memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
+}
+
+/*
  * Log key values from the btree block.
  */
 STATIC void
@@ -1163,6 +1234,79 @@ xfs_btree_log_recs(
 }
 
 /*
+ * Log block pointer fields from a btree block (nonleaf).
+ */
+STATIC void
+xfs_btree_log_ptrs(
+	struct xfs_btree_cur	*cur,	/* btree cursor */
+	struct xfs_buf		*bp,	/* buffer containing btree block */
+	int			first,	/* index of first pointer to log */
+	int			last)	/* index of last pointer to log */
+{
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+	XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
+
+	if (bp) {
+		struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
+		int			level = xfs_btree_get_level(block);
+
+		xfs_trans_log_buf(cur->bc_tp, bp,
+				xfs_btree_ptr_offset(cur, first, level),
+				xfs_btree_ptr_offset(cur, last + 1, level) - 1);
+	} else {
+		xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
+			xfs_ilog_fbroot(cur->bc_private.b.whichfork));
+	}
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+}
+
+/*
+ * Log fields from a btree block header.
+ */
+STATIC void
+xfs_btree_log_block(
+	struct xfs_btree_cur	*cur,	/* btree cursor */
+	struct xfs_buf		*bp,	/* buffer containing btree block */
+	int			fields)	/* mask of fields: XFS_BB_... */
+{
+	int			first;	/* first byte offset logged */
+	int			last;	/* last byte offset logged */
+	static const short	soffsets[] = {	/* table of offsets (short) */
+		offsetof(struct xfs_btree_sblock, bb_magic),
+		offsetof(struct xfs_btree_sblock, bb_level),
+		offsetof(struct xfs_btree_sblock, bb_numrecs),
+		offsetof(struct xfs_btree_sblock, bb_leftsib),
+		offsetof(struct xfs_btree_sblock, bb_rightsib),
+		sizeof(struct xfs_btree_sblock)
+	};
+	static const short	loffsets[] = {	/* table of offsets (long) */
+		offsetof(struct xfs_btree_lblock, bb_magic),
+		offsetof(struct xfs_btree_lblock, bb_level),
+		offsetof(struct xfs_btree_lblock, bb_numrecs),
+		offsetof(struct xfs_btree_lblock, bb_leftsib),
+		offsetof(struct xfs_btree_lblock, bb_rightsib),
+		sizeof(struct xfs_btree_lblock)
+	};
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+	XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
+
+	if (bp) {
+		xfs_btree_offsets(fields,
+				  (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
+					loffsets : soffsets,
+				  XFS_BB_NUM_BITS, &first, &last);
+		xfs_trans_log_buf(cur->bc_tp, bp, first, last);
+	} else {
+		xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
+			xfs_ilog_fbroot(cur->bc_private.b.whichfork));
+	}
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+}
+
+/*
  * Increment cursor by one record at the level.
  * For nonzero levels the leaf-ward information is untouched.
  */
@@ -1368,7 +1512,6 @@ error0:
 	return error;
 }
 
-
 STATIC int
 xfs_btree_lookup_get_block(
 	struct xfs_btree_cur	*cur,	/* btree cursor */
@@ -1697,3 +1840,177 @@ error0:
 	return error;
 }
 
+/*
+ * Move 1 record right from cur/level if possible.
+ * Update cur to reflect the new path.
+ */
+int					/* error */
+xfs_btree_rshift(
+	struct xfs_btree_cur	*cur,
+	int			level,
+	int			*stat)		/* success/failure */
+{
+	union xfs_btree_key	key;		/* btree key */
+	struct xfs_buf		*lbp;		/* left buffer pointer */
+	struct xfs_btree_block	*left;		/* left btree block */
+	struct xfs_buf		*rbp;		/* right buffer pointer */
+	struct xfs_btree_block	*right;		/* right btree block */
+	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
+	union xfs_btree_ptr	rptr;		/* right block pointer */
+	union xfs_btree_key	*rkp;		/* right btree key */
+	int			rrecs;		/* right record count */
+	int			lrecs;		/* left record count */
+	int			error;		/* error return value */
+	int			i;		/* loop counter */
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+	XFS_BTREE_TRACE_ARGI(cur, level);
+
+	if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
+	    (level == cur->bc_nlevels - 1))
+		goto out0;
+
+	/* Set up variables for this block as "left". */
+	left = xfs_btree_get_block(cur, level, &lbp);
+
+#ifdef DEBUG
+	error = xfs_btree_check_block(cur, left, level, lbp);
+	if (error)
+		goto error0;
+#endif
+
+	/* If we've got no right sibling then we can't shift an entry right. */
+	xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
+	if (xfs_btree_ptr_is_null(cur, &rptr))
+		goto out0;
+
+	/*
+	 * If the cursor entry is the one that would be moved, don't
+	 * do it... it's too complicated.
+	 */
+	lrecs = xfs_btree_get_numrecs(left);
+	if (cur->bc_ptrs[level] >= lrecs)
+		goto out0;
+
+	/* Set up the right neighbor as "right". */
+	error = xfs_btree_read_buf_block(cur, &rptr, level, 0, &right, &rbp);
+	if (error)
+		goto error0;
+
+	/* If it's full, it can't take another entry. */
+	rrecs = xfs_btree_get_numrecs(right);
+	if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
+		goto out0;
+
+	XFS_BTREE_STATS_INC(cur, rshift);
+	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
+
+	/*
+	 * Make a hole at the start of the right neighbor block, then
+	 * copy the last left block entry to the hole.
+	 */
+	if (level > 0) {
+		/* It's a nonleaf. make a hole in the keys and ptrs */
+		union xfs_btree_key	*lkp;
+		union xfs_btree_ptr	*lpp;
+		union xfs_btree_ptr	*rpp;
+
+		lkp = xfs_btree_key_addr(cur, lrecs, left);
+		lpp = xfs_btree_ptr_addr(cur, lrecs, left);
+		rkp = xfs_btree_key_addr(cur, 1, right);
+		rpp = xfs_btree_ptr_addr(cur, 1, right);
+
+#ifdef DEBUG
+		for (i = rrecs - 1; i >= 0; i--) {
+			error = xfs_btree_check_ptr(cur, rpp, i, level);
+			if (error)
+				goto error0;
+		}
+#endif
+
+		xfs_btree_shift_keys(cur, rkp, 1, rrecs);
+		xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
+
+#ifdef DEBUG
+		error = xfs_btree_check_ptr(cur, lpp, 0, level);
+		if (error)
+			goto error0;
+#endif
+
+		/* Now put the new data in, and log it. */
+		xfs_btree_copy_keys(cur, rkp, lkp, 1);
+		xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
+
+		xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
+		xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
+
+		xfs_btree_check_key(cur->bc_btnum, rkp,
+				    xfs_btree_key_addr(cur, 2, right));
+	} else {
+		/* It's a leaf. make a hole in the records */
+		union xfs_btree_rec	*lrp;
+		union xfs_btree_rec	*rrp;
+
+		lrp = xfs_btree_rec_addr(cur, lrecs, left);
+		rrp = xfs_btree_rec_addr(cur, 1, right);
+
+		xfs_btree_shift_recs(cur, rrp, 1, rrecs);
+
+		/* Now put the new data in, and log it. */
+		xfs_btree_copy_recs(cur, rrp, lrp, 1);
+		xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
+
+		cur->bc_ops->init_key_from_rec(&key, rrp);
+		rkp = &key;
+
+		xfs_btree_check_rec(cur->bc_btnum, rrp,
+				    xfs_btree_rec_addr(cur, 2, right));
+	}
+
+	/*
+	 * Decrement and log left's numrecs, bump and log right's numrecs.
+	 */
+	xfs_btree_set_numrecs(left, --lrecs);
+	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
+
+	xfs_btree_set_numrecs(right, ++rrecs);
+	xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
+
+	/*
+	 * Using a temporary cursor, update the parent key values of the
+	 * block on the right.
+	 */
+	error = xfs_btree_dup_cursor(cur, &tcur);
+	if (error)
+		goto error0;
+	i = xfs_btree_lastrec(tcur, level);
+	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
+
+	error = xfs_btree_increment(tcur, level, &i);
+	if (error)
+		goto error1;
+
+	error = xfs_btree_updkey(tcur, rkp, level + 1);
+	if (error)
+		goto error1;
+
+	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
+
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+	*stat = 1;
+	return 0;
+
+out0:
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+	*stat = 0;
+	return 0;
+
+error0:
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+	return error;
+
+error1:
+	XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
+	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
+	return error;
+}
diff --git a/fs/xfs/xfs_btree.h b/fs/xfs/xfs_btree.h
index c3bfa5556c1..04311dbeff1 100644
--- a/fs/xfs/xfs_btree.h
+++ b/fs/xfs/xfs_btree.h
@@ -533,6 +533,7 @@ int xfs_btree_decrement(struct xfs_btree_cur *, int, int *);
 int xfs_btree_lookup(struct xfs_btree_cur *, xfs_lookup_t, int *);
 int xfs_btree_updkey(struct xfs_btree_cur *, union xfs_btree_key *, int);
 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
+int xfs_btree_rshift(struct xfs_btree_cur *, int, int *);
 
 /*
  * Helpers.
@@ -542,6 +543,12 @@ static inline int xfs_btree_get_numrecs(struct xfs_btree_block *block)
 	return be16_to_cpu(block->bb_numrecs);
 }
 
+static inline void xfs_btree_set_numrecs(struct xfs_btree_block *block,
+		__uint16_t numrecs)
+{
+	block->bb_numrecs = cpu_to_be16(numrecs);
+}
+
 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
 {
 	return be16_to_cpu(block->bb_level);
diff --git a/fs/xfs/xfs_ialloc_btree.c b/fs/xfs/xfs_ialloc_btree.c
index d080a6833a8..457f88a76e1 100644
--- a/fs/xfs/xfs_ialloc_btree.c
+++ b/fs/xfs/xfs_ialloc_btree.c
@@ -45,7 +45,6 @@ STATIC void xfs_inobt_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_inobt_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC int xfs_inobt_lshift(xfs_btree_cur_t *, int, int *);
 STATIC int xfs_inobt_newroot(xfs_btree_cur_t *, int *);
-STATIC int xfs_inobt_rshift(xfs_btree_cur_t *, int, int *);
 STATIC int xfs_inobt_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
 		xfs_inobt_key_t *, xfs_btree_cur_t **, int *);
 
@@ -337,7 +336,7 @@ xfs_inobt_delrec(
 		 */
 		if (be16_to_cpu(left->bb_numrecs) - 1 >=
 		     XFS_INOBT_BLOCK_MINRECS(level, cur)) {
-			if ((error = xfs_inobt_rshift(tcur, level, &i)))
+			if ((error = xfs_btree_rshift(tcur, level, &i)))
 				goto error0;
 			if (i) {
 				ASSERT(be16_to_cpu(block->bb_numrecs) >=
@@ -608,7 +607,7 @@ xfs_inobt_insrec(
 		/*
 		 * First, try shifting an entry to the right neighbor.
 		 */
-		if ((error = xfs_inobt_rshift(cur, level, &i)))
+		if ((error = xfs_btree_rshift(cur, level, &i)))
 			return error;
 		if (i) {
 			/* nothing */
@@ -1117,136 +1116,6 @@ xfs_inobt_newroot(
 }
 
 /*
- * Move 1 record right from cur/level if possible.
- * Update cur to reflect the new path.
- */
-STATIC int				/* error */
-xfs_inobt_rshift(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	int			level,	/* level to shift record on */
-	int			*stat)	/* success/failure */
-{
-	int			error;	/* error return value */
-	int			i;	/* loop index */
-	xfs_inobt_key_t		key;	/* key value for leaf level upward */
-	xfs_buf_t		*lbp;	/* buffer for left (current) block */
-	xfs_inobt_block_t	*left;	/* left (current) btree block */
-	xfs_inobt_key_t		*lkp;	/* key pointer for left block */
-	xfs_inobt_ptr_t		*lpp;	/* address pointer for left block */
-	xfs_inobt_rec_t		*lrp;	/* record pointer for left block */
-	xfs_buf_t		*rbp;	/* buffer for right neighbor block */
-	xfs_inobt_block_t	*right;	/* right neighbor btree block */
-	xfs_inobt_key_t		*rkp;	/* key pointer for right block */
-	xfs_inobt_ptr_t		*rpp;	/* address pointer for right block */
-	xfs_inobt_rec_t		*rrp=NULL;	/* record pointer for right block */
-	xfs_btree_cur_t		*tcur;	/* temporary cursor */
-
-	/*
-	 * Set up variables for this block as "left".
-	 */
-	lbp = cur->bc_bufs[level];
-	left = XFS_BUF_TO_INOBT_BLOCK(lbp);
-#ifdef DEBUG
-	if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-		return error;
-#endif
-	/*
-	 * If we've got no right sibling then we can't shift an entry right.
-	 */
-	if (be32_to_cpu(left->bb_rightsib) == NULLAGBLOCK) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * If the cursor entry is the one that would be moved, don't
-	 * do it... it's too complicated.
-	 */
-	if (cur->bc_ptrs[level] >= be16_to_cpu(left->bb_numrecs)) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * Set up the right neighbor as "right".
-	 */
-	if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
-			cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib),
-			0, &rbp, XFS_INO_BTREE_REF)))
-		return error;
-	right = XFS_BUF_TO_INOBT_BLOCK(rbp);
-	if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
-		return error;
-	/*
-	 * If it's full, it can't take another entry.
-	 */
-	if (be16_to_cpu(right->bb_numrecs) == XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * Make a hole at the start of the right neighbor block, then
-	 * copy the last left block entry to the hole.
-	 */
-	if (level > 0) {
-		lkp = XFS_INOBT_KEY_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		lpp = XFS_INOBT_PTR_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
-		rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
-#ifdef DEBUG
-		for (i = be16_to_cpu(right->bb_numrecs) - 1; i >= 0; i--) {
-			if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
-				return error;
-		}
-#endif
-		memmove(rkp + 1, rkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-		memmove(rpp + 1, rpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-#ifdef DEBUG
-		if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(*lpp), level)))
-			return error;
-#endif
-		*rkp = *lkp;
-		*rpp = *lpp;
-		xfs_inobt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		xfs_inobt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-	} else {
-		lrp = XFS_INOBT_REC_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
-		rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
-		memmove(rrp + 1, rrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-		*rrp = *lrp;
-		xfs_inobt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
-		key.ir_startino = rrp->ir_startino;
-		rkp = &key;
-	}
-	/*
-	 * Decrement and log left's numrecs, bump and log right's numrecs.
-	 */
-	be16_add_cpu(&left->bb_numrecs, -1);
-	xfs_inobt_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS);
-	be16_add_cpu(&right->bb_numrecs, 1);
-#ifdef DEBUG
-	if (level > 0)
-		xfs_btree_check_key(cur->bc_btnum, rkp, rkp + 1);
-	else
-		xfs_btree_check_rec(cur->bc_btnum, rrp, rrp + 1);
-#endif
-	xfs_inobt_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS);
-	/*
-	 * Using a temporary cursor, update the parent key values of the
-	 * block on the right.
-	 */
-	if ((error = xfs_btree_dup_cursor(cur, &tcur)))
-		return error;
-	xfs_btree_lastrec(tcur, level);
-	if ((error = xfs_btree_increment(tcur, level, &i)) ||
-	    (error = xfs_btree_updkey(tcur, (union xfs_btree_key *)rkp, level + 1))) {
-		xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
-		return error;
-	}
-	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
-	*stat = 1;
-	return 0;
-}
-
-/*
  * Split cur/level block in half.
  * Return new block number and its first record (to be inserted into parent).
  */