Merge tag 'for-5.19-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs reverts from David Sterba:
 "Due to a recent report [1] we need to revert the radix tree to xarray
  conversion patches.

  There's a problem with sleeping under spinlock, when xa_insert could
  allocate memory under pressure. We use GFP_NOFS so this is a real
  problem that we unfortunately did not discover during review.

  I'm sorry to do such change at rc6 time but the revert is IMO the
  safer option, there are patches to use mutex instead of the spin locks
  but that would need more testing. The revert branch has been tested on
  a few setups, all seem ok.

  The conversion to xarray will be revisited in the future"

Link: https://lore.kernel.org/linux-btrfs/cover.1657097693.git.fdmanana@suse.com/ [1]

* tag 'for-5.19-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  Revert "btrfs: turn delayed_nodes_tree into an XArray"
  Revert "btrfs: turn name_cache radix tree into XArray in send_ctx"
  Revert "btrfs: turn fs_info member buffer_radix into XArray"
  Revert "btrfs: turn fs_roots_radix in btrfs_fs_info into an XArray"

Author: torvalds

fs/btrfs/ctree.h

@@ -675,9 +675,8 @@ struct btrfs_fs_info {
 	rwlock_t global_root_lock;
 	struct rb_root global_root_tree;
 
-	/* The xarray that holds all the FS roots */
-	spinlock_t fs_roots_lock;
-	struct xarray fs_roots;
+	spinlock_t fs_roots_radix_lock;
+	struct radix_tree_root fs_roots_radix;
 
 	/* block group cache stuff */
 	rwlock_t block_group_cache_lock;
@@ -995,10 +994,10 @@ struct btrfs_fs_info {
 
 	struct btrfs_delayed_root *delayed_root;
 
-	/* Extent buffer xarray */
+	/* Extent buffer radix tree */
 	spinlock_t buffer_lock;
 	/* Entries are eb->start / sectorsize */
-	struct xarray extent_buffers;
+	struct radix_tree_root buffer_radix;
 
 	/* next backup root to be overwritten */
 	int backup_root_index;
@@ -1119,8 +1118,7 @@ enum {
 	 */
 	BTRFS_ROOT_SHAREABLE,
 	BTRFS_ROOT_TRACK_DIRTY,
-	/* The root is tracked in fs_info::fs_roots */
-	BTRFS_ROOT_REGISTERED,
+	BTRFS_ROOT_IN_RADIX,
 	BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
 	BTRFS_ROOT_DEFRAG_RUNNING,
 	BTRFS_ROOT_FORCE_COW,
@@ -1224,10 +1222,10 @@ struct btrfs_root {
 	struct rb_root inode_tree;
 
 	/*
-	 * Xarray that keeps track of delayed nodes of every inode, protected
-	 * by inode_lock
+	 * radix tree that keeps track of delayed nodes of every inode,
+	 * protected by inode_lock
 	 */
-	struct xarray delayed_nodes;
+	struct radix_tree_root delayed_nodes_tree;
 	/*
 	 * right now this just gets used so that a root has its own devid
 	 * for stat.  It may be used for more later

fs/btrfs/delayed-inode.c

@@ -78,7 +78,7 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 	}
 
 	spin_lock(&root->inode_lock);
-	node = xa_load(&root->delayed_nodes, ino);
+	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 
 	if (node) {
 		if (btrfs_inode->delayed_node) {
@@ -90,17 +90,17 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 
 		/*
 		 * It's possible that we're racing into the middle of removing
-		 * this node from the xarray.  In this case, the refcount
+		 * this node from the radix tree.  In this case, the refcount
 		 * was zero and it should never go back to one.  Just return
-		 * NULL like it was never in the xarray at all; our release
+		 * NULL like it was never in the radix at all; our release
 		 * function is in the process of removing it.
 		 *
 		 * Some implementations of refcount_inc refuse to bump the
 		 * refcount once it has hit zero.  If we don't do this dance
 		 * here, refcount_inc() may decide to just WARN_ONCE() instead
 		 * of actually bumping the refcount.
 		 *
-		 * If this node is properly in the xarray, we want to bump the
+		 * If this node is properly in the radix, we want to bump the
 		 * refcount twice, once for the inode and once for this get
 		 * operation.
 		 */
@@ -128,30 +128,36 @@ static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 	u64 ino = btrfs_ino(btrfs_inode);
 	int ret;
 
-	do {
-		node = btrfs_get_delayed_node(btrfs_inode);
-		if (node)
-			return node;
+again:
+	node = btrfs_get_delayed_node(btrfs_inode);
+	if (node)
+		return node;
 
-		node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
-		if (!node)
-			return ERR_PTR(-ENOMEM);
-		btrfs_init_delayed_node(node, root, ino);
+	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
+	if (!node)
+		return ERR_PTR(-ENOMEM);
+	btrfs_init_delayed_node(node, root, ino);
 
-		/* Cached in the inode and can be accessed */
-		refcount_set(&node->refs, 2);
+	/* cached in the btrfs inode and can be accessed */
+	refcount_set(&node->refs, 2);
 
-		spin_lock(&root->inode_lock);
-		ret = xa_insert(&root->delayed_nodes, ino, node, GFP_NOFS);
-		if (ret) {
-			spin_unlock(&root->inode_lock);
-			kmem_cache_free(delayed_node_cache, node);
-			if (ret != -EBUSY)
-				return ERR_PTR(ret);
-		}
-	} while (ret);
+	ret = radix_tree_preload(GFP_NOFS);
+	if (ret) {
+		kmem_cache_free(delayed_node_cache, node);
+		return ERR_PTR(ret);
+	}
+
+	spin_lock(&root->inode_lock);
+	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
+	if (ret == -EEXIST) {
+		spin_unlock(&root->inode_lock);
+		kmem_cache_free(delayed_node_cache, node);
+		radix_tree_preload_end();
+		goto again;
+	}
 	btrfs_inode->delayed_node = node;
 	spin_unlock(&root->inode_lock);
+	radix_tree_preload_end();
 
 	return node;
 }
@@ -270,7 +276,8 @@ static void __btrfs_release_delayed_node(
 		 * back up.  We can delete it now.
 		 */
 		ASSERT(refcount_read(&delayed_node->refs) == 0);
-		xa_erase(&root->delayed_nodes, delayed_node->inode_id);
+		radix_tree_delete(&root->delayed_nodes_tree,
+				  delayed_node->inode_id);
 		spin_unlock(&root->inode_lock);
 		kmem_cache_free(delayed_node_cache, delayed_node);
 	}
@@ -1863,35 +1870,34 @@ void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
 
 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
 {
-	unsigned long index = 0;
-	struct btrfs_delayed_node *delayed_node;
+	u64 inode_id = 0;
 	struct btrfs_delayed_node *delayed_nodes[8];
+	int i, n;
 
 	while (1) {
-		int n = 0;
-
 		spin_lock(&root->inode_lock);
-		if (xa_empty(&root->delayed_nodes)) {
+		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
+					   (void **)delayed_nodes, inode_id,
+					   ARRAY_SIZE(delayed_nodes));
+		if (!n) {
 			spin_unlock(&root->inode_lock);
-			return;
+			break;
 		}
 
-		xa_for_each_start(&root->delayed_nodes, index, delayed_node, index) {
+		inode_id = delayed_nodes[n - 1]->inode_id + 1;
+		for (i = 0; i < n; i++) {
 			/*
 			 * Don't increase refs in case the node is dead and
 			 * about to be removed from the tree in the loop below
 			 */
-			if (refcount_inc_not_zero(&delayed_node->refs)) {
-				delayed_nodes[n] = delayed_node;
-				n++;
-			}
-			if (n >= ARRAY_SIZE(delayed_nodes))
-				break;
+			if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
+				delayed_nodes[i] = NULL;
 		}
-		index++;
 		spin_unlock(&root->inode_lock);
 
-		for (int i = 0; i < n; i++) {
+		for (i = 0; i < n; i++) {
+			if (!delayed_nodes[i])
+				continue;
 			__btrfs_kill_delayed_node(delayed_nodes[i]);
 			btrfs_release_delayed_node(delayed_nodes[i]);
 		}