1 files changed, 88 insertions, 26 deletions

diff --git a/fs/xfs/libxfs/xfs_btree.c b/fs/xfs/libxfs/xfs_btree.c
index 5d0b1084cba8..769be61ad63f 100644
--- a/fs/xfs/libxfs/xfs_btree.c
+++ b/fs/xfs/libxfs/xfs_btree.c
@@ -298,17 +298,17 @@ xfs_btree_check_ptr(
 				level))
 			return 0;
 		xfs_err(cur->bc_mp,
-"Inode %llu fork %d: Corrupt btree %d pointer at level %d index %d.",
+"Inode %llu fork %d: Corrupt %sbt pointer at level %d index %d.",
 				cur->bc_ino.ip->i_ino,
-				cur->bc_ino.whichfork, cur->bc_btnum,
+				cur->bc_ino.whichfork, cur->bc_ops->name,
 				level, index);
 	} else {
 		if (xfs_btree_check_sptr(cur, be32_to_cpu((&ptr->s)[index]),
 				level))
 			return 0;
 		xfs_err(cur->bc_mp,
-"AG %u: Corrupt btree %d pointer at level %d index %d.",
-				cur->bc_ag.pag->pag_agno, cur->bc_btnum,
+"AG %u: Corrupt %sbt pointer at level %d index %d.",
+				cur->bc_ag.pag->pag_agno, cur->bc_ops->name,
 				level, index);
 	}
 
@@ -407,6 +407,15 @@ xfs_btree_free_block(
 
 	trace_xfs_btree_free_block(cur, bp);
 
+	/*
+	 * Don't allow block freeing for a staging cursor, because staging
+	 * cursors do not support regular btree modifications.
+	 */
+	if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
+		ASSERT(0);
+		return -EFSCORRUPTED;
+	}
+
 	error = cur->bc_ops->free_block(cur, bp);
 	if (!error) {
 		xfs_trans_binval(cur->bc_tp, bp);
@@ -445,7 +454,7 @@ xfs_btree_del_cursor(
 	 * zero, then we should be shut down or on our way to shutdown due to
 	 * cancelling a dirty transaction on error.
 	 */
-	ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP || cur->bc_bmap.allocated == 0 ||
+	ASSERT(!xfs_btree_is_bmap(cur->bc_ops) || cur->bc_bmap.allocated == 0 ||
 	       xfs_is_shutdown(cur->bc_mp) || error != 0);
 
 	switch (cur->bc_ops->type) {
@@ -458,8 +467,6 @@ xfs_btree_del_cursor(
 		break;
 	}
 
-	if (unlikely(cur->bc_flags & XFS_BTREE_STAGING))
-		kfree(cur->bc_ops);
 	kmem_cache_free(cur->bc_cache, cur);
 }
 
@@ -467,20 +474,26 @@ xfs_btree_del_cursor(
  * Duplicate the btree cursor.
  * Allocate a new one, copy the record, re-get the buffers.
  */
-int					/* error */
+int						/* error */
 xfs_btree_dup_cursor(
-	struct xfs_btree_cur *cur,		/* input cursor */
-	struct xfs_btree_cur **ncur)		/* output cursor */
+	struct xfs_btree_cur	*cur,		/* input cursor */
+	struct xfs_btree_cur	**ncur)		/* output cursor */
 {
-	struct xfs_buf	*bp;		/* btree block's buffer pointer */
-	int		error;		/* error return value */
-	int		i;		/* level number of btree block */
-	xfs_mount_t	*mp;		/* mount structure for filesystem */
-	struct xfs_btree_cur *new;		/* new cursor value */
-	xfs_trans_t	*tp;		/* transaction pointer, can be NULL */
+	struct xfs_mount	*mp = cur->bc_mp;
+	struct xfs_trans	*tp = cur->bc_tp;
+	struct xfs_buf		*bp;
+	struct xfs_btree_cur	*new;
+	int			error;
+	int			i;
 
-	tp = cur->bc_tp;
-	mp = cur->bc_mp;
+	/*
+	 * Don't allow staging cursors to be duplicated because they're supposed
+	 * to be kept private to a single thread.
+	 */
+	if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
+		ASSERT(0);
+		return -EFSCORRUPTED;
+	}
 
 	/*
 	 * Allocate a new cursor like the old one.
@@ -1882,6 +1895,27 @@ xfs_lookup_get_search_key(
 }
 
 /*
+ * Initialize a pointer to the root block.
+ */
+void
+xfs_btree_init_ptr_from_cur(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*ptr)
+{
+	if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE) {
+		/*
+		 * Inode-rooted btrees call xfs_btree_get_iroot to find the root
+		 * in xfs_btree_lookup_get_block and don't need a pointer here.
+		 */
+		ptr->l = 0;
+	} else if (cur->bc_flags & XFS_BTREE_STAGING) {
+		ptr->s = cpu_to_be32(cur->bc_ag.afake->af_root);
+	} else {
+		cur->bc_ops->init_ptr_from_cur(cur, ptr);
+	}
+}
+
+/*
  * Lookup the record.  The cursor is made to point to it, based on dir.
  * stat is set to 0 if can't find any such record, 1 for success.
  */
@@ -1911,7 +1945,7 @@ xfs_btree_lookup(
 	keyno = 0;
 
 	/* initialise start pointer from cursor */
-	cur->bc_ops->init_ptr_from_cur(cur, &ptr);
+	xfs_btree_init_ptr_from_cur(cur, &ptr);
 	pp = &ptr;
 
 	/*
@@ -2697,6 +2731,18 @@ xfs_btree_alloc_block(
 {
 	int				error;
 
+	/*
+	 * Don't allow block allocation for a staging cursor, because staging
+	 * cursors do not support regular btree modifications.
+	 *
+	 * Bulk loading uses a separate callback to obtain new blocks from a
+	 * preallocated list, which prevents ENOSPC failures during loading.
+	 */
+	if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
+		ASSERT(0);
+		return -EFSCORRUPTED;
+	}
+
 	error = cur->bc_ops->alloc_block(cur, hint_block, new_block, stat);
 	trace_xfs_btree_alloc_block(cur, new_block, *stat, error);
 	return error;
@@ -2970,7 +3016,7 @@ xfs_btree_split(
 	struct xfs_btree_split_args	args;
 	DECLARE_COMPLETION_ONSTACK(done);
 
-	if (cur->bc_btnum != XFS_BTNUM_BMAP ||
+	if (!xfs_btree_is_bmap(cur->bc_ops) ||
 	    cur->bc_tp->t_highest_agno == NULLAGNUMBER)
 		return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
 
@@ -3097,6 +3143,21 @@ error0:
 	return error;
 }
 
+static void
+xfs_btree_set_root(
+	struct xfs_btree_cur		*cur,
+	const union xfs_btree_ptr	*ptr,
+	int				inc)
+{
+	if (cur->bc_flags & XFS_BTREE_STAGING) {
+		/* Update the btree root information for a per-AG fake root. */
+		cur->bc_ag.afake->af_root = be32_to_cpu(ptr->s);
+		cur->bc_ag.afake->af_levels += inc;
+	} else {
+		cur->bc_ops->set_root(cur, ptr, inc);
+	}
+}
+
 /*
  * Allocate a new root block, fill it in.
  */
@@ -3121,7 +3182,7 @@ xfs_btree_new_root(
 	XFS_BTREE_STATS_INC(cur, newroot);
 
 	/* initialise our start point from the cursor */
-	cur->bc_ops->init_ptr_from_cur(cur, &rptr);
+	xfs_btree_init_ptr_from_cur(cur, &rptr);
 
 	/* Allocate the new block. If we can't do it, we're toast. Give up. */
 	error = xfs_btree_alloc_block(cur, &rptr, &lptr, stat);
@@ -3137,7 +3198,7 @@ xfs_btree_new_root(
 		goto error0;
 
 	/* Set the root in the holding structure  increasing the level by 1. */
-	cur->bc_ops->set_root(cur, &lptr, 1);
+	xfs_btree_set_root(cur, &lptr, 1);
 
 	/*
 	 * At the previous root level there are now two blocks: the old root,
@@ -3565,7 +3626,8 @@ xfs_btree_insert(
 		if (pcur != cur &&
 		    (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
 			/* Save the state from the cursor before we trash it */
-			if (cur->bc_ops->update_cursor)
+			if (cur->bc_ops->update_cursor &&
+			    !(cur->bc_flags & XFS_BTREE_STAGING))
 				cur->bc_ops->update_cursor(pcur, cur);
 			cur->bc_nlevels = pcur->bc_nlevels;
 			xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
@@ -3708,7 +3770,7 @@ xfs_btree_kill_root(
 	 * Update the root pointer, decreasing the level by 1 and then
 	 * free the old root.
 	 */
-	cur->bc_ops->set_root(cur, newroot, -1);
+	xfs_btree_set_root(cur, newroot, -1);
 
 	error = xfs_btree_free_block(cur, bp);
 	if (error)
@@ -4430,7 +4492,7 @@ xfs_btree_visit_blocks(
 	struct xfs_btree_block		*block = NULL;
 	int				error = 0;
 
-	cur->bc_ops->init_ptr_from_cur(cur, &lptr);
+	xfs_btree_init_ptr_from_cur(cur, &lptr);
 
 	/* for each level */
 	for (level = cur->bc_nlevels - 1; level >= 0; level--) {
@@ -4852,7 +4914,7 @@ xfs_btree_overlapped_query_range(
 
 	/* Load the root of the btree. */
 	level = cur->bc_nlevels - 1;
-	cur->bc_ops->init_ptr_from_cur(cur, &ptr);
+	xfs_btree_init_ptr_from_cur(cur, &ptr);
 	error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
 	if (error)
 		return error;