1 files changed, 249 insertions, 1944 deletions

diff --git a/fs/xfs/xfs_ialloc_btree.c b/fs/xfs/xfs_ialloc_btree.c
index 83502f3edef..99f2408e8d8 100644
--- a/fs/xfs/xfs_ialloc_btree.c
+++ b/fs/xfs/xfs_ialloc_btree.c
@@ -35,2044 +35,349 @@
 #include "xfs_dinode.h"
 #include "xfs_inode.h"
 #include "xfs_btree.h"
+#include "xfs_btree_trace.h"
 #include "xfs_ialloc.h"
 #include "xfs_alloc.h"
 #include "xfs_error.h"
 
-STATIC void xfs_inobt_log_block(xfs_trans_t *, xfs_buf_t *, int);
-STATIC void xfs_inobt_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
-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 *);
-STATIC int xfs_inobt_updkey(xfs_btree_cur_t *, xfs_inobt_key_t *, int);
 
-/*
- * Single level of the xfs_inobt_delete record deletion routine.
- * Delete record pointed to by cur/level.
- * Remove the record from its block then rebalance the tree.
- * Return 0 for error, 1 for done, 2 to go on to the next level.
- */
-STATIC int				/* error */
-xfs_inobt_delrec(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	int			level,	/* level removing record from */
-	int			*stat)	/* fail/done/go-on */
+STATIC int
+xfs_inobt_get_minrecs(
+	struct xfs_btree_cur	*cur,
+	int			level)
 {
-	xfs_buf_t		*agbp;	/* buffer for a.g. inode header */
-	xfs_mount_t		*mp;	/* mount structure */
-	xfs_agi_t		*agi;	/* allocation group inode header */
-	xfs_inobt_block_t	*block;	/* btree block record/key lives in */
-	xfs_agblock_t		bno;	/* btree block number */
-	xfs_buf_t		*bp;	/* buffer for block */
-	int			error;	/* error return value */
-	int			i;	/* loop index */
-	xfs_inobt_key_t		key;	/* kp points here if block is level 0 */
-	xfs_inobt_key_t		*kp = NULL;	/* pointer to btree keys */
-	xfs_agblock_t		lbno;	/* left block's block number */
-	xfs_buf_t		*lbp;	/* left block's buffer pointer */
-	xfs_inobt_block_t	*left;	/* left btree block */
-	xfs_inobt_key_t		*lkp;	/* left block key pointer */
-	xfs_inobt_ptr_t		*lpp;	/* left block address pointer */
-	int			lrecs = 0;	/* number of records in left block */
-	xfs_inobt_rec_t		*lrp;	/* left block record pointer */
-	xfs_inobt_ptr_t		*pp = NULL;	/* pointer to btree addresses */
-	int			ptr;	/* index in btree block for this rec */
-	xfs_agblock_t		rbno;	/* right block's block number */
-	xfs_buf_t		*rbp;	/* right block's buffer pointer */
-	xfs_inobt_block_t	*right;	/* right btree block */
-	xfs_inobt_key_t		*rkp;	/* right block key pointer */
-	xfs_inobt_rec_t		*rp;	/* pointer to btree records */
-	xfs_inobt_ptr_t		*rpp;	/* right block address pointer */
-	int			rrecs = 0;	/* number of records in right block */
-	int			numrecs;
-	xfs_inobt_rec_t		*rrp;	/* right block record pointer */
-	xfs_btree_cur_t		*tcur;	/* temporary btree cursor */
-
-	mp = cur->bc_mp;
-
-	/*
-	 * Get the index of the entry being deleted, check for nothing there.
-	 */
-	ptr = cur->bc_ptrs[level];
-	if (ptr == 0) {
-		*stat = 0;
-		return 0;
-	}
-
-	/*
-	 * Get the buffer & block containing the record or key/ptr.
-	 */
-	bp = cur->bc_bufs[level];
-	block = XFS_BUF_TO_INOBT_BLOCK(bp);
-#ifdef DEBUG
-	if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
-		return error;
-#endif
-	/*
-	 * Fail if we're off the end of the block.
-	 */
+	return cur->bc_mp->m_inobt_mnr[level != 0];
+}
 
-	numrecs = be16_to_cpu(block->bb_numrecs);
-	if (ptr > numrecs) {
-		*stat = 0;
-		return 0;
-	}
-	/*
-	 * It's a nonleaf.  Excise the key and ptr being deleted, by
-	 * sliding the entries past them down one.
-	 * Log the changed areas of the block.
-	 */
-	if (level > 0) {
-		kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
-		pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
-#ifdef DEBUG
-		for (i = ptr; i < numrecs; i++) {
-			if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(pp[i]), level)))
-				return error;
-		}
-#endif
-		if (ptr < numrecs) {
-			memmove(&kp[ptr - 1], &kp[ptr],
-				(numrecs - ptr) * sizeof(*kp));
-			memmove(&pp[ptr - 1], &pp[ptr],
-				(numrecs - ptr) * sizeof(*kp));
-			xfs_inobt_log_keys(cur, bp, ptr, numrecs - 1);
-			xfs_inobt_log_ptrs(cur, bp, ptr, numrecs - 1);
-		}
-	}
-	/*
-	 * It's a leaf.  Excise the record being deleted, by sliding the
-	 * entries past it down one.  Log the changed areas of the block.
-	 */
-	else {
-		rp = XFS_INOBT_REC_ADDR(block, 1, cur);
-		if (ptr < numrecs) {
-			memmove(&rp[ptr - 1], &rp[ptr],
-				(numrecs - ptr) * sizeof(*rp));
-			xfs_inobt_log_recs(cur, bp, ptr, numrecs - 1);
-		}
-		/*
-		 * If it's the first record in the block, we'll need a key
-		 * structure to pass up to the next level (updkey).
-		 */
-		if (ptr == 1) {
-			key.ir_startino = rp->ir_startino;
-			kp = &key;
-		}
-	}
-	/*
-	 * Decrement and log the number of entries in the block.
-	 */
-	numrecs--;
-	block->bb_numrecs = cpu_to_be16(numrecs);
-	xfs_inobt_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
-	/*
-	 * Is this the root level?  If so, we're almost done.
-	 */
-	if (level == cur->bc_nlevels - 1) {
-		/*
-		 * If this is the root level,
-		 * and there's only one entry left,
-		 * and it's NOT the leaf level,
-		 * then we can get rid of this level.
-		 */
-		if (numrecs == 1 && level > 0) {
-			agbp = cur->bc_private.a.agbp;
-			agi = XFS_BUF_TO_AGI(agbp);
-			/*
-			 * pp is still set to the first pointer in the block.
-			 * Make it the new root of the btree.
-			 */
-			bno = be32_to_cpu(agi->agi_root);
-			agi->agi_root = *pp;
-			be32_add_cpu(&agi->agi_level, -1);
-			/*
-			 * Free the block.
-			 */
-			if ((error = xfs_free_extent(cur->bc_tp,
-				XFS_AGB_TO_FSB(mp, cur->bc_private.a.agno, bno), 1)))
-				return error;
-			xfs_trans_binval(cur->bc_tp, bp);
-			xfs_ialloc_log_agi(cur->bc_tp, agbp,
-				XFS_AGI_ROOT | XFS_AGI_LEVEL);
-			/*
-			 * Update the cursor so there's one fewer level.
-			 */
-			cur->bc_bufs[level] = NULL;
-			cur->bc_nlevels--;
-		} else if (level > 0 &&
-			   (error = xfs_inobt_decrement(cur, level, &i)))
-			return error;
-		*stat = 1;
-		return 0;
-	}
-	/*
-	 * If we deleted the leftmost entry in the block, update the
-	 * key values above us in the tree.
-	 */
-	if (ptr == 1 && (error = xfs_inobt_updkey(cur, kp, level + 1)))
-		return error;
-	/*
-	 * If the number of records remaining in the block is at least
-	 * the minimum, we're done.
-	 */
-	if (numrecs >= XFS_INOBT_BLOCK_MINRECS(level, cur)) {
-		if (level > 0 &&
-		    (error = xfs_inobt_decrement(cur, level, &i)))
-			return error;
-		*stat = 1;
-		return 0;
-	}
-	/*
-	 * Otherwise, we have to move some records around to keep the
-	 * tree balanced.  Look at the left and right sibling blocks to
-	 * see if we can re-balance by moving only one record.
-	 */
-	rbno = be32_to_cpu(block->bb_rightsib);
-	lbno = be32_to_cpu(block->bb_leftsib);
-	bno = NULLAGBLOCK;
-	ASSERT(rbno != NULLAGBLOCK || lbno != NULLAGBLOCK);
-	/*
-	 * Duplicate the cursor so our btree manipulations here won't
-	 * disrupt the next level up.
-	 */
-	if ((error = xfs_btree_dup_cursor(cur, &tcur)))
-		return error;
-	/*
-	 * If there's a right sibling, see if it's ok to shift an entry
-	 * out of it.
-	 */
-	if (rbno != NULLAGBLOCK) {
-		/*
-		 * Move the temp cursor to the last entry in the next block.
-		 * Actually any entry but the first would suffice.
-		 */
-		i = xfs_btree_lastrec(tcur, level);
-		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-		if ((error = xfs_inobt_increment(tcur, level, &i)))
-			goto error0;
-		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-		i = xfs_btree_lastrec(tcur, level);
-		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
-		/*
-		 * Grab a pointer to the block.
-		 */
-		rbp = tcur->bc_bufs[level];
-		right = XFS_BUF_TO_INOBT_BLOCK(rbp);
-#ifdef DEBUG
-		if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
-			goto error0;
-#endif
-		/*
-		 * Grab the current block number, for future use.
-		 */
-		bno = be32_to_cpu(right->bb_leftsib);
-		/*
-		 * If right block is full enough so that removing one entry
-		 * won't make it too empty, and left-shifting an entry out
-		 * of right to us works, we're done.
-		 */
-		if (be16_to_cpu(right->bb_numrecs) - 1 >=
-		     XFS_INOBT_BLOCK_MINRECS(level, cur)) {
-			if ((error = xfs_inobt_lshift(tcur, level, &i)))
-				goto error0;
-			if (i) {
-				ASSERT(be16_to_cpu(block->bb_numrecs) >=
-				       XFS_INOBT_BLOCK_MINRECS(level, cur));
-				xfs_btree_del_cursor(tcur,
-						     XFS_BTREE_NOERROR);
-				if (level > 0 &&
-				    (error = xfs_inobt_decrement(cur, level,
-						&i)))
-					return error;
-				*stat = 1;
-				return 0;
-			}
-		}
-		/*
-		 * Otherwise, grab the number of records in right for
-		 * future reference, and fix up the temp cursor to point
-		 * to our block again (last record).
-		 */
-		rrecs = be16_to_cpu(right->bb_numrecs);
-		if (lbno != NULLAGBLOCK) {
-			xfs_btree_firstrec(tcur, level);
-			if ((error = xfs_inobt_decrement(tcur, level, &i)))
-				goto error0;
-		}
-	}
-	/*
-	 * If there's a left sibling, see if it's ok to shift an entry
-	 * out of it.
-	 */
-	if (lbno != NULLAGBLOCK) {
-		/*
-		 * Move the temp cursor to the first entry in the
-		 * previous block.
-		 */
-		xfs_btree_firstrec(tcur, level);
-		if ((error = xfs_inobt_decrement(tcur, level, &i)))
-			goto error0;
-		xfs_btree_firstrec(tcur, level);
-		/*
-		 * Grab a pointer to the block.
-		 */
-		lbp = tcur->bc_bufs[level];
-		left = XFS_BUF_TO_INOBT_BLOCK(lbp);
-#ifdef DEBUG
-		if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-			goto error0;
-#endif
-		/*
-		 * Grab the current block number, for future use.
-		 */
-		bno = be32_to_cpu(left->bb_rightsib);
-		/*
-		 * If left block is full enough so that removing one entry
-		 * won't make it too empty, and right-shifting an entry out
-		 * of left to us works, we're done.
-		 */
-		if (be16_to_cpu(left->bb_numrecs) - 1 >=
-		     XFS_INOBT_BLOCK_MINRECS(level, cur)) {
-			if ((error = xfs_inobt_rshift(tcur, level, &i)))
-				goto error0;
-			if (i) {
-				ASSERT(be16_to_cpu(block->bb_numrecs) >=
-				       XFS_INOBT_BLOCK_MINRECS(level, cur));
-				xfs_btree_del_cursor(tcur,
-						     XFS_BTREE_NOERROR);
-				if (level == 0)
-					cur->bc_ptrs[0]++;
-				*stat = 1;
-				return 0;
-			}
-		}
-		/*
-		 * Otherwise, grab the number of records in right for
-		 * future reference.
-		 */
-		lrecs = be16_to_cpu(left->bb_numrecs);
-	}
-	/*
-	 * Delete the temp cursor, we're done with it.
-	 */
-	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
-	/*
-	 * If here, we need to do a join to keep the tree balanced.
-	 */
-	ASSERT(bno != NULLAGBLOCK);
-	/*
-	 * See if we can join with the left neighbor block.
-	 */
-	if (lbno != NULLAGBLOCK &&
-	    lrecs + numrecs <= XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
-		/*
-		 * Set "right" to be the starting block,
-		 * "left" to be the left neighbor.
-		 */
-		rbno = bno;
-		right = block;
-		rrecs = be16_to_cpu(right->bb_numrecs);
-		rbp = bp;
-		if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
-				cur->bc_private.a.agno, lbno, 0, &lbp,
-				XFS_INO_BTREE_REF)))
-			return error;
-		left = XFS_BUF_TO_INOBT_BLOCK(lbp);
-		lrecs = be16_to_cpu(left->bb_numrecs);
-		if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-			return error;
-	}
-	/*
-	 * If that won't work, see if we can join with the right neighbor block.
-	 */
-	else if (rbno != NULLAGBLOCK &&
-		 rrecs + numrecs <= XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
-		/*
-		 * Set "left" to be the starting block,
-		 * "right" to be the right neighbor.
-		 */
-		lbno = bno;
-		left = block;
-		lrecs = be16_to_cpu(left->bb_numrecs);
-		lbp = bp;
-		if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
-				cur->bc_private.a.agno, rbno, 0, &rbp,
-				XFS_INO_BTREE_REF)))
-			return error;
-		right = XFS_BUF_TO_INOBT_BLOCK(rbp);
-		rrecs = be16_to_cpu(right->bb_numrecs);
-		if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
-			return error;
-	}
-	/*
-	 * Otherwise, we can't fix the imbalance.
-	 * Just return.  This is probably a logic error, but it's not fatal.
-	 */
-	else {
-		if (level > 0 && (error = xfs_inobt_decrement(cur, level, &i)))
-			return error;
-		*stat = 1;
-		return 0;
-	}
-	/*
-	 * We're now going to join "left" and "right" by moving all the stuff
-	 * in "right" to "left" and deleting "right".
-	 */
-	if (level > 0) {
-		/*
-		 * It's a non-leaf.  Move keys and pointers.
-		 */
-		lkp = XFS_INOBT_KEY_ADDR(left, lrecs + 1, cur);
-		lpp = XFS_INOBT_PTR_ADDR(left, lrecs + 1, cur);
-		rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
-		rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
-#ifdef DEBUG
-		for (i = 0; i < rrecs; i++) {
-			if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
-				return error;
-		}
-#endif
-		memcpy(lkp, rkp, rrecs * sizeof(*lkp));
-		memcpy(lpp, rpp, rrecs * sizeof(*lpp));
-		xfs_inobt_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
-		xfs_inobt_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
-	} else {
-		/*
-		 * It's a leaf.  Move records.
-		 */
-		lrp = XFS_INOBT_REC_ADDR(left, lrecs + 1, cur);
-		rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
-		memcpy(lrp, rrp, rrecs * sizeof(*lrp));
-		xfs_inobt_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
-	}
-	/*
-	 * If we joined with the left neighbor, set the buffer in the
-	 * cursor to the left block, and fix up the index.
-	 */
-	if (bp != lbp) {
-		xfs_btree_setbuf(cur, level, lbp);
-		cur->bc_ptrs[level] += lrecs;
-	}
-	/*
-	 * If we joined with the right neighbor and there's a level above
-	 * us, increment the cursor at that level.
-	 */
-	else if (level + 1 < cur->bc_nlevels &&
-		 (error = xfs_alloc_increment(cur, level + 1, &i)))
-		return error;
-	/*
-	 * Fix up the number of records in the surviving block.
-	 */
-	lrecs += rrecs;
-	left->bb_numrecs = cpu_to_be16(lrecs);
-	/*
-	 * Fix up the right block pointer in the surviving block, and log it.
-	 */
-	left->bb_rightsib = right->bb_rightsib;
-	xfs_inobt_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
-	/*
-	 * If there is a right sibling now, make it point to the
-	 * remaining block.
-	 */
-	if (be32_to_cpu(left->bb_rightsib) != NULLAGBLOCK) {
-		xfs_inobt_block_t	*rrblock;
-		xfs_buf_t		*rrbp;
+STATIC struct xfs_btree_cur *
+xfs_inobt_dup_cursor(
+	struct xfs_btree_cur	*cur)
+{
+	return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
+			cur->bc_private.a.agbp, cur->bc_private.a.agno);
+}
 
-		if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
-				cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib), 0,
-				&rrbp, XFS_INO_BTREE_REF)))
-			return error;
-		rrblock = XFS_BUF_TO_INOBT_BLOCK(rrbp);
-		if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
-			return error;
-		rrblock->bb_leftsib = cpu_to_be32(lbno);
-		xfs_inobt_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB);
-	}
-	/*
-	 * Free the deleting block.
-	 */
-	if ((error = xfs_free_extent(cur->bc_tp, XFS_AGB_TO_FSB(mp,
-				     cur->bc_private.a.agno, rbno), 1)))
-		return error;
-	xfs_trans_binval(cur->bc_tp, rbp);
-	/*
-	 * Readjust the ptr at this level if it's not a leaf, since it's
-	 * still pointing at the deletion point, which makes the cursor
-	 * inconsistent.  If this makes the ptr 0, the caller fixes it up.
-	 * We can't use decrement because it would change the next level up.
-	 */
-	if (level > 0)
-		cur->bc_ptrs[level]--;
-	/*
-	 * Return value means the next level up has something to do.
-	 */
-	*stat = 2;
-	return 0;
+STATIC void
+xfs_inobt_set_root(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*nptr,
+	int			inc)	/* level change */
+{
+	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
+	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
 
-error0:
-	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
-	return error;
+	agi->agi_root = nptr->s;
+	be32_add_cpu(&agi->agi_level, inc);
+	xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
 }
 
-/*
- * Insert one record/level.  Return information to the caller
- * allowing the next level up to proceed if necessary.
- */
-STATIC int				/* error */
-xfs_inobt_insrec(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	int			level,	/* level to insert record at */
-	xfs_agblock_t		*bnop,	/* i/o: block number inserted */
-	xfs_inobt_rec_t		*recp,	/* i/o: record data inserted */
-	xfs_btree_cur_t		**curp,	/* output: new cursor replacing cur */
-	int			*stat)	/* success/failure */
+STATIC int
+xfs_inobt_alloc_block(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*start,
+	union xfs_btree_ptr	*new,
+	int			length,
+	int			*stat)
 {
-	xfs_inobt_block_t	*block;	/* btree block record/key lives in */
-	xfs_buf_t		*bp;	/* buffer for block */
-	int			error;	/* error return value */
-	int			i;	/* loop index */
-	xfs_inobt_key_t		key;	/* key value being inserted */
-	xfs_inobt_key_t		*kp=NULL;	/* pointer to btree keys */
-	xfs_agblock_t		nbno;	/* block number of allocated block */
-	xfs_btree_cur_t		*ncur;	/* new cursor to be used at next lvl */
-	xfs_inobt_key_t		nkey;	/* new key value, from split */
-	xfs_inobt_rec_t		nrec;	/* new record value, for caller */
-	int			numrecs;
-	int			optr;	/* old ptr value */
-	xfs_inobt_ptr_t		*pp;	/* pointer to btree addresses */
-	int			ptr;	/* index in btree block for this rec */
-	xfs_inobt_rec_t		*rp=NULL;	/* pointer to btree records */
+	xfs_alloc_arg_t		args;		/* block allocation args */
+	int			error;		/* error return value */
+	xfs_agblock_t		sbno = be32_to_cpu(start->s);
 
-	/*
-	 * GCC doesn't understand the (arguably complex) control flow in
-	 * this function and complains about uninitialized structure fields
-	 * without this.
-	 */
-	memset(&nrec, 0, sizeof(nrec));
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
 
-	/*
-	 * If we made it to the root level, allocate a new root block
-	 * and we're done.
-	 */
-	if (level >= cur->bc_nlevels) {
-		error = xfs_inobt_newroot(cur, &i);
-		*bnop = NULLAGBLOCK;
-		*stat = i;
+	memset(&args, 0, sizeof(args));
+	args.tp = cur->bc_tp;
+	args.mp = cur->bc_mp;
+	args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_private.a.agno, sbno);
+	args.minlen = 1;
+	args.maxlen = 1;
+	args.prod = 1;
+	args.type = XFS_ALLOCTYPE_NEAR_BNO;
+
+	error = xfs_alloc_vextent(&args);
+	if (error) {
+		XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
 		return error;
 	}
-	/*
-	 * Make a key out of the record data to be inserted, and save it.
-	 */
-	key.ir_startino = recp->ir_startino;
-	optr = ptr = cur->bc_ptrs[level];
-	/*
-	 * If we're off the left edge, return failure.
-	 */
-	if (ptr == 0) {
+	if (args.fsbno == NULLFSBLOCK) {
+		XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
 		*stat = 0;
 		return 0;
 	}
-	/*
-	 * Get pointers to the btree buffer and block.
-	 */
-	bp = cur->bc_bufs[level];
-	block = XFS_BUF_TO_INOBT_BLOCK(bp);
-	numrecs = be16_to_cpu(block->bb_numrecs);
-#ifdef DEBUG
-	if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
-		return error;
-	/*
-	 * Check that the new entry is being inserted in the right place.
-	 */
-	if (ptr <= numrecs) {
-		if (level == 0) {
-			rp = XFS_INOBT_REC_ADDR(block, ptr, cur);
-			xfs_btree_check_rec(cur->bc_btnum, recp, rp);
-		} else {
-			kp = XFS_INOBT_KEY_ADDR(block, ptr, cur);
-			xfs_btree_check_key(cur->bc_btnum, &key, kp);
-		}
-	}
-#endif
-	nbno = NULLAGBLOCK;
-	ncur = NULL;
-	/*
-	 * If the block is full, we can't insert the new entry until we
-	 * make the block un-full.
-	 */
-	if (numrecs == XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
-		/*
-		 * First, try shifting an entry to the right neighbor.
-		 */
-		if ((error = xfs_inobt_rshift(cur, level, &i)))
-			return error;
-		if (i) {
-			/* nothing */
-		}
-		/*
-		 * Next, try shifting an entry to the left neighbor.
-		 */
-		else {
-			if ((error = xfs_inobt_lshift(cur, level, &i)))
-				return error;
-			if (i) {
-				optr = ptr = cur->bc_ptrs[level];
-			} else {
-				/*
-				 * Next, try splitting the current block
-				 * in half. If this works we have to
-				 * re-set our variables because
-				 * we could be in a different block now.
-				 */
-				if ((error = xfs_inobt_split(cur, level, &nbno,
-						&nkey, &ncur, &i)))
-					return error;
-				if (i) {
-					bp = cur->bc_bufs[level];
-					block = XFS_BUF_TO_INOBT_BLOCK(bp);
-#ifdef DEBUG
-					if ((error = xfs_btree_check_sblock(cur,
-							block, level, bp)))
-						return error;
-#endif
-					ptr = cur->bc_ptrs[level];
-					nrec.ir_startino = nkey.ir_startino;
-				} else {
-					/*
-					 * Otherwise the insert fails.
-					 */
-					*stat = 0;
-					return 0;
-				}
-			}
-		}
-	}
-	/*
-	 * At this point we know there's room for our new entry in the block
-	 * we're pointing at.
-	 */
-	numrecs = be16_to_cpu(block->bb_numrecs);
-	if (level > 0) {
-		/*
-		 * It's a non-leaf entry.  Make a hole for the new data
-		 * in the key and ptr regions of the block.
-		 */
-		kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
-		pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
-#ifdef DEBUG
-		for (i = numrecs; i >= ptr; i--) {
-			if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(pp[i - 1]), level)))
-				return error;
-		}
-#endif
-		memmove(&kp[ptr], &kp[ptr - 1],
-			(numrecs - ptr + 1) * sizeof(*kp));
-		memmove(&pp[ptr], &pp[ptr - 1],
-			(numrecs - ptr + 1) * sizeof(*pp));
-		/*
-		 * Now stuff the new data in, bump numrecs and log the new data.
-		 */
-#ifdef DEBUG
-		if ((error = xfs_btree_check_sptr(cur, *bnop, level)))
-			return error;
-#endif
-		kp[ptr - 1] = key;
-		pp[ptr - 1] = cpu_to_be32(*bnop);
-		numrecs++;
-		block->bb_numrecs = cpu_to_be16(numrecs);
-		xfs_inobt_log_keys(cur, bp, ptr, numrecs);
-		xfs_inobt_log_ptrs(cur, bp, ptr, numrecs);
-	} else {
-		/*
-		 * It's a leaf entry.  Make a hole for the new record.
-		 */
-		rp = XFS_INOBT_REC_ADDR(block, 1, cur);
-		memmove(&rp[ptr], &rp[ptr - 1],
-			(numrecs - ptr + 1) * sizeof(*rp));
-		/*
-		 * Now stuff the new record in, bump numrecs
-		 * and log the new data.
-		 */
-		rp[ptr - 1] = *recp;
-		numrecs++;
-		block->bb_numrecs = cpu_to_be16(numrecs);
-		xfs_inobt_log_recs(cur, bp, ptr, numrecs);
-	}
-	/*
-	 * Log the new number of records in the btree header.
-	 */
-	xfs_inobt_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
-#ifdef DEBUG
-	/*
-	 * Check that the key/record is in the right place, now.
-	 */
-	if (ptr < numrecs) {
-		if (level == 0)
-			xfs_btree_check_rec(cur->bc_btnum, rp + ptr - 1,
-				rp + ptr);
-		else
-			xfs_btree_check_key(cur->bc_btnum, kp + ptr - 1,
-				kp + ptr);
-	}
-#endif
-	/*
-	 * If we inserted at the start of a block, update the parents' keys.
-	 */
-	if (optr == 1 && (error = xfs_inobt_updkey(cur, &key, level + 1)))
-		return error;
-	/*
-	 * Return the new block number, if any.
-	 * If there is one, give back a record value and a cursor too.
-	 */
-	*bnop = nbno;
-	if (nbno != NULLAGBLOCK) {
-		*recp = nrec;
-		*curp = ncur;
-	}
+	ASSERT(args.len == 1);
+	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+
+	new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
 	*stat = 1;
 	return 0;
 }
 
-/*
- * Log header fields from a btree block.
- */
-STATIC void
-xfs_inobt_log_block(
-	xfs_trans_t		*tp,	/* transaction pointer */
-	xfs_buf_t		*bp,	/* buffer containing btree block */
-	int			fields)	/* mask of fields: XFS_BB_... */
+STATIC int
+xfs_inobt_free_block(
+	struct xfs_btree_cur	*cur,
+	struct xfs_buf		*bp)
 {
-	int			first;	/* first byte offset logged */
-	int			last;	/* last byte offset logged */
-	static const short	offsets[] = {	/* table of offsets */
-		offsetof(xfs_inobt_block_t, bb_magic),
-		offsetof(xfs_inobt_block_t, bb_level),
-		offsetof(xfs_inobt_block_t, bb_numrecs),
-		offsetof(xfs_inobt_block_t, bb_leftsib),
-		offsetof(xfs_inobt_block_t, bb_rightsib),
-		sizeof(xfs_inobt_block_t)
-	};
+	xfs_fsblock_t		fsbno;
+	int			error;
 
-	xfs_btree_offsets(fields, offsets, XFS_BB_NUM_BITS, &first, &last);
-	xfs_trans_log_buf(tp, bp, first, last);
+	fsbno = XFS_DADDR_TO_FSB(cur->bc_mp, XFS_BUF_ADDR(bp));
+	error = xfs_free_extent(cur->bc_tp, fsbno, 1);
+	if (error)
+		return error;
+
+	xfs_trans_binval(cur->bc_tp, bp);
+	return error;
 }
 
-/*
- * Log keys from a btree block (nonleaf).
- */
-STATIC void
-xfs_inobt_log_keys(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	xfs_buf_t		*bp,	/* buffer containing btree block */
-	int			kfirst,	/* index of first key to log */
-	int			klast)	/* index of last key to log */
+STATIC int
+xfs_inobt_get_maxrecs(
+	struct xfs_btree_cur	*cur,
+	int			level)
 {
-	xfs_inobt_block_t	*block;	/* btree block to log from */
-	int			first;	/* first byte offset logged */
-	xfs_inobt_key_t		*kp;	/* key pointer in btree block */
-	int			last;	/* last byte offset logged */
-
-	block = XFS_BUF_TO_INOBT_BLOCK(bp);
-	kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
-	first = (int)((xfs_caddr_t)&kp[kfirst - 1] - (xfs_caddr_t)block);
-	last = (int)(((xfs_caddr_t)&kp[klast] - 1) - (xfs_caddr_t)block);
-	xfs_trans_log_buf(cur->bc_tp, bp, first, last);
+	return cur->bc_mp->m_inobt_mxr[level != 0];
 }
 
-/*
- * Log block pointer fields from a btree block (nonleaf).
- */
 STATIC void
-xfs_inobt_log_ptrs(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	xfs_buf_t		*bp,	/* buffer containing btree block */
-	int			pfirst,	/* index of first pointer to log */
-	int			plast)	/* index of last pointer to log */
+xfs_inobt_init_key_from_rec(
+	union xfs_btree_key	*key,
+	union xfs_btree_rec	*rec)
 {
-	xfs_inobt_block_t	*block;	/* btree block to log from */
-	int			first;	/* first byte offset logged */
-	int			last;	/* last byte offset logged */
-	xfs_inobt_ptr_t		*pp;	/* block-pointer pointer in btree blk */
-
-	block = XFS_BUF_TO_INOBT_BLOCK(bp);
-	pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
-	first = (int)((xfs_caddr_t)&pp[pfirst - 1] - (xfs_caddr_t)block);
-	last = (int)(((xfs_caddr_t)&pp[plast] - 1) - (xfs_caddr_t)block);
-	xfs_trans_log_buf(cur->bc_tp, bp, first, last);
+	key->inobt.ir_startino = rec->inobt.ir_startino;
 }
 
-/*
- * Log records from a btree block (leaf).
- */
 STATIC void
-xfs_inobt_log_recs(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	xfs_buf_t		*bp,	/* buffer containing btree block */
-	int			rfirst,	/* index of first record to log */
-	int			rlast)	/* index of last record to log */
+xfs_inobt_init_rec_from_key(
+	union xfs_btree_key	*key,
+	union xfs_btree_rec	*rec)
 {
-	xfs_inobt_block_t	*block;	/* btree block to log from */
-	int			first;	/* first byte offset logged */
-	int			last;	/* last byte offset logged */
-	xfs_inobt_rec_t		*rp;	/* record pointer for btree block */
+	rec->inobt.ir_startino = key->inobt.ir_startino;
+}
 
-	block = XFS_BUF_TO_INOBT_BLOCK(bp);
-	rp = XFS_INOBT_REC_ADDR(block, 1, cur);
-	first = (int)((xfs_caddr_t)&rp[rfirst - 1] - (xfs_caddr_t)block);
-	last = (int)(((xfs_caddr_t)&rp[rlast] - 1) - (xfs_caddr_t)block);
-	xfs_trans_log_buf(cur->bc_tp, bp, first, last);
+STATIC void
+xfs_inobt_init_rec_from_cur(
+	struct xfs_btree_cur	*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);
+	rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
 }
 
 /*
- * Lookup the record.  The cursor is made to point to it, based on dir.
- * Return 0 if can't find any such record, 1 for success.
+ * intial value of ptr for lookup
  */
-STATIC int				/* error */
-xfs_inobt_lookup(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	xfs_lookup_t		dir,	/* <=, ==, or >= */
-	int			*stat)	/* success/failure */
+STATIC void
+xfs_inobt_init_ptr_from_cur(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*ptr)
 {
-	xfs_agblock_t		agbno;	/* a.g. relative btree block number */
-	xfs_agnumber_t		agno;	/* allocation group number */
-	xfs_inobt_block_t	*block=NULL;	/* current btree block */
-	__int64_t		diff;	/* difference for the current key */
-	int			error;	/* error return value */
-	int			keyno=0;	/* current key number */
-	int			level;	/* level in the btree */
-	xfs_mount_t		*mp;	/* file system mount point */
-
-	/*
-	 * Get the allocation group header, and the root block number.
-	 */
-	mp = cur->bc_mp;
-	{
-		xfs_agi_t	*agi;	/* a.g. inode header */
-
-		agi = XFS_BUF_TO_AGI(cur->bc_private.a.agbp);
-		agno = be32_to_cpu(agi->agi_seqno);
-		agbno = be32_to_cpu(agi->agi_root);
-	}
-	/*
-	 * Iterate over each level in the btree, starting at the root.
-	 * For each level above the leaves, find the key we need, based
-	 * on the lookup record, then follow the corresponding block
-	 * pointer down to the next level.
-	 */
-	for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
-		xfs_buf_t	*bp;	/* buffer pointer for btree block */
-		xfs_daddr_t	d;	/* disk address of btree block */
-
-		/*
-		 * Get the disk address we're looking for.
-		 */
-		d = XFS_AGB_TO_DADDR(mp, agno, agbno);
-		/*
-		 * If the old buffer at this level is for a different block,
-		 * throw it away, otherwise just use it.
-		 */
-		bp = cur->bc_bufs[level];
-		if (bp && XFS_BUF_ADDR(bp) != d)
-			bp = NULL;
-		if (!bp) {
-			/*
-			 * Need to get a new buffer.  Read it, then
-			 * set it in the cursor, releasing the old one.
-			 */
-			if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
-					agno, agbno, 0, &bp, XFS_INO_BTREE_REF)))
-				return error;
-			xfs_btree_setbuf(cur, level, bp);
-			/*
-			 * Point to the btree block, now that we have the buffer
-			 */
-			block = XFS_BUF_TO_INOBT_BLOCK(bp);
-			if ((error = xfs_btree_check_sblock(cur, block, level,
-					bp)))
-				return error;
-		} else
-			block = XFS_BUF_TO_INOBT_BLOCK(bp);
-		/*
-		 * If we already had a key match at a higher level, we know
-		 * we need to use the first entry in this block.
-		 */
-		if (diff == 0)
-			keyno = 1;
-		/*
-		 * Otherwise we need to search this block.  Do a binary search.
-		 */
-		else {
-			int		high;	/* high entry number */
-			xfs_inobt_key_t	*kkbase=NULL;/* base of keys in block */
-			xfs_inobt_rec_t	*krbase=NULL;/* base of records in block */
-			int		low;	/* low entry number */
+	struct xfs_agi		*agi = XFS_BUF_TO_AGI(cur->bc_private.a.agbp);
 
-			/*
-			 * Get a pointer to keys or records.
-			 */
-			if (level > 0)
-				kkbase = XFS_INOBT_KEY_ADDR(block, 1, cur);
-			else
-				krbase = XFS_INOBT_REC_ADDR(block, 1, cur);
-			/*
-			 * Set low and high entry numbers, 1-based.
-			 */
-			low = 1;
-			if (!(high = be16_to_cpu(block->bb_numrecs))) {
-				/*
-				 * If the block is empty, the tree must
-				 * be an empty leaf.
-				 */
-				ASSERT(level == 0 && cur->bc_nlevels == 1);
-				cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
-				*stat = 0;
-				return 0;
-			}
-			/*
-			 * Binary search the block.
-			 */
-			while (low <= high) {
-				xfs_agino_t	startino;	/* key value */
-
-				/*
-				 * keyno is average of low and high.
-				 */
-				keyno = (low + high) >> 1;
-				/*
-				 * Get startino.
-				 */
-				if (level > 0) {
-					xfs_inobt_key_t	*kkp;
-
-					kkp = kkbase + keyno - 1;
-					startino = be32_to_cpu(kkp->ir_startino);
-				} else {
-					xfs_inobt_rec_t	*krp;
-
-					krp = krbase + keyno - 1;
-					startino = be32_to_cpu(krp->ir_startino);
-				}
-				/*
-				 * Compute difference to get next direction.
-				 */
-				diff = (__int64_t)
-					startino - cur->bc_rec.i.ir_startino;
-				/*
-				 * Less than, move right.
-				 */
-				if (diff < 0)
-					low = keyno + 1;
-				/*
-				 * Greater than, move left.
-				 */
-				else if (diff > 0)
-					high = keyno - 1;
-				/*
-				 * Equal, we're done.
-				 */
-				else
-					break;
-			}
-		}
-		/*
-		 * If there are more levels, set up for the next level
-		 * by getting the block number and filling in the cursor.
-		 */
-		if (level > 0) {
-			/*
-			 * If we moved left, need the previous key number,
-			 * unless there isn't one.
-			 */
-			if (diff > 0 && --keyno < 1)
-				keyno = 1;
-			agbno = be32_to_cpu(*XFS_INOBT_PTR_ADDR(block, keyno, cur));
-#ifdef DEBUG
-			if ((error = xfs_btree_check_sptr(cur, agbno, level)))
-				return error;
-#endif
-			cur->bc_ptrs[level] = keyno;
-		}
-	}
-	/*
-	 * Done with the search.
-	 * See if we need to adjust the results.
-	 */
-	if (dir != XFS_LOOKUP_LE && diff < 0) {
-		keyno++;
-		/*
-		 * If ge search and we went off the end of the block, but it's
-		 * not the last block, we're in the wrong block.
-		 */
-		if (dir == XFS_LOOKUP_GE &&
-		    keyno > be16_to_cpu(block->bb_numrecs) &&
-		    be32_to_cpu(block->bb_rightsib) != NULLAGBLOCK) {
-			int	i;
+	ASSERT(cur->bc_private.a.agno == be32_to_cpu(agi->agi_seqno));
 
-			cur->bc_ptrs[0] = keyno;
-			if ((error = xfs_inobt_increment(cur, 0, &i)))
-				return error;
-			ASSERT(i == 1);
-			*stat = 1;
-			return 0;
-		}
-	}
-	else if (dir == XFS_LOOKUP_LE && diff > 0)
-		keyno--;
-	cur->bc_ptrs[0] = keyno;
-	/*
-	 * Return if we succeeded or not.
-	 */
-	if (keyno == 0 || keyno > be16_to_cpu(block->bb_numrecs))
-		*stat = 0;
-	else
-		*stat = ((dir != XFS_LOOKUP_EQ) || (diff == 0));
-	return 0;
+	ptr->s = agi->agi_root;
 }
 
-/*
- * Move 1 record left from cur/level if possible.
- * Update cur to reflect the new path.
- */
-STATIC int				/* error */
-xfs_inobt_lshift(
-	xfs_btree_cur_t		*cur,	/* btree cursor */
-	int			level,	/* level to shift record on */
-	int			*stat)	/* success/failure */
+STATIC __int64_t
+xfs_inobt_key_diff(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_key	*key)
 {
-	int			error;	/* error return value */
-#ifdef DEBUG
-	int			i;	/* loop index */
-#endif
-	xfs_inobt_key_t		key;	/* key value for leaf level upward */
-	xfs_buf_t		*lbp;	/* buffer for left neighbor block */
-	xfs_inobt_block_t	*left;	/* left neighbor btree block */
-	xfs_inobt_key_t		*lkp=NULL;	/* key pointer for left block */
-	xfs_inobt_ptr_t		*lpp;	/* address pointer for left block */
-	xfs_inobt_rec_t		*lrp=NULL;	/* record pointer for left block */
-	int			nrec;	/* new number of left block entries */
-	xfs_buf_t		*rbp;	/* buffer for right (current) block */
-	xfs_inobt_block_t	*right;	/* right (current) btree block */
-	xfs_inobt_key_t		*rkp=NULL;	/* key pointer for right block */
-	xfs_inobt_ptr_t		*rpp=NULL;	/* address pointer for right block */
-	xfs_inobt_rec_t		*rrp=NULL;	/* record pointer for right block */
-
-	/*
-	 * Set up variables for this block as "right".
-	 */
-	rbp = cur-&g