Revert "btrfs: turn fs_roots_radix in btrfs_fs_info into an XArray"

This reverts commit 48b36a602a335c184505346b5b37077840660634.

Revert the xarray conversion, there's a problem with potential
sleep-inside-spinlock [1] when calling xa_insert that triggers GFP_NOFS
allocation. The radix tree used the preloading mechanism to avoid
sleeping but this is not available in xarray.

Conversion from spin lock to mutex is possible but at time of rc6 is
riskier than a clean revert.

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

Reported-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 63 insertions, 49 deletions

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 06c0a958d114..875b801ab3d7 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -23,7 +23,7 @@
 #include "space-info.h"
 #include "zoned.h"
 
-#define BTRFS_ROOT_TRANS_TAG				XA_MARK_0
+#define BTRFS_ROOT_TRANS_TAG 0
 
 /*
  * Transaction states and transitions
@@ -437,15 +437,15 @@ static int record_root_in_trans(struct btrfs_trans_handle *trans,
 		 */
 		smp_wmb();
 
-		spin_lock(&fs_info->fs_roots_lock);
+		spin_lock(&fs_info->fs_roots_radix_lock);
 		if (root->last_trans == trans->transid && !force) {
-			spin_unlock(&fs_info->fs_roots_lock);
+			spin_unlock(&fs_info->fs_roots_radix_lock);
 			return 0;
 		}
-		xa_set_mark(&fs_info->fs_roots,
-			    (unsigned long)root->root_key.objectid,
-			    BTRFS_ROOT_TRANS_TAG);
-		spin_unlock(&fs_info->fs_roots_lock);
+		radix_tree_tag_set(&fs_info->fs_roots_radix,
+				   (unsigned long)root->root_key.objectid,
+				   BTRFS_ROOT_TRANS_TAG);
+		spin_unlock(&fs_info->fs_roots_radix_lock);
 		root->last_trans = trans->transid;
 
 		/* this is pretty tricky.  We don't want to
@@ -487,9 +487,11 @@ void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
 	spin_unlock(&cur_trans->dropped_roots_lock);
 
 	/* Make sure we don't try to update the root at commit time */
-	xa_clear_mark(&fs_info->fs_roots,
-		      (unsigned long)root->root_key.objectid,
-		      BTRFS_ROOT_TRANS_TAG);
+	spin_lock(&fs_info->fs_roots_radix_lock);
+	radix_tree_tag_clear(&fs_info->fs_roots_radix,
+			     (unsigned long)root->root_key.objectid,
+			     BTRFS_ROOT_TRANS_TAG);
+	spin_unlock(&fs_info->fs_roots_radix_lock);
 }
 
 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
@@ -1402,8 +1404,9 @@ void btrfs_add_dead_root(struct btrfs_root *root)
 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
-	struct btrfs_root *root;
-	unsigned long index;
+	struct btrfs_root *gang[8];
+	int i;
+	int ret;
 
 	/*
 	 * At this point no one can be using this transaction to modify any tree
@@ -1411,46 +1414,57 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 	 */
 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
 
-	spin_lock(&fs_info->fs_roots_lock);
-	xa_for_each_marked(&fs_info->fs_roots, index, root, BTRFS_ROOT_TRANS_TAG) {
-		int ret;
-
-		/*
-		 * At this point we can neither have tasks logging inodes
-		 * from a root nor trying to commit a log tree.
-		 */
-		ASSERT(atomic_read(&root->log_writers) == 0);
-		ASSERT(atomic_read(&root->log_commit[0]) == 0);
-		ASSERT(atomic_read(&root->log_commit[1]) == 0);
-
-		xa_clear_mark(&fs_info->fs_roots,
-			      (unsigned long)root->root_key.objectid,
-			      BTRFS_ROOT_TRANS_TAG);
-		spin_unlock(&fs_info->fs_roots_lock);
-
-		btrfs_free_log(trans, root);
-		ret = btrfs_update_reloc_root(trans, root);
-		if (ret)
-			return ret;
-
-		/* See comments in should_cow_block() */
-		clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
-		smp_mb__after_atomic();
+	spin_lock(&fs_info->fs_roots_radix_lock);
+	while (1) {
+		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
+						 (void **)gang, 0,
+						 ARRAY_SIZE(gang),
+						 BTRFS_ROOT_TRANS_TAG);
+		if (ret == 0)
+			break;
+		for (i = 0; i < ret; i++) {
+			struct btrfs_root *root = gang[i];
+			int ret2;
+
+			/*
+			 * At this point we can neither have tasks logging inodes
+			 * from a root nor trying to commit a log tree.
+			 */
+			ASSERT(atomic_read(&root->log_writers) == 0);
+			ASSERT(atomic_read(&root->log_commit[0]) == 0);
+			ASSERT(atomic_read(&root->log_commit[1]) == 0);
+
+			radix_tree_tag_clear(&fs_info->fs_roots_radix,
+					(unsigned long)root->root_key.objectid,
+					BTRFS_ROOT_TRANS_TAG);
+			spin_unlock(&fs_info->fs_roots_radix_lock);
+
+			btrfs_free_log(trans, root);
+			ret2 = btrfs_update_reloc_root(trans, root);
+			if (ret2)
+				return ret2;
+
+			/* see comments in should_cow_block() */
+			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
+			smp_mb__after_atomic();
+
+			if (root->commit_root != root->node) {
+				list_add_tail(&root->dirty_list,
+					&trans->transaction->switch_commits);
+				btrfs_set_root_node(&root->root_item,
+						    root->node);
+			}
 
-		if (root->commit_root != root->node) {
-			list_add_tail(&root->dirty_list,
-				      &trans->transaction->switch_commits);
-			btrfs_set_root_node(&root->root_item, root->node);
+			ret2 = btrfs_update_root(trans, fs_info->tree_root,
+						&root->root_key,
+						&root->root_item);
+			if (ret2)
+				return ret2;
+			spin_lock(&fs_info->fs_roots_radix_lock);
+			btrfs_qgroup_free_meta_all_pertrans(root);
 		}
-
-		ret = btrfs_update_root(trans, fs_info->tree_root,
-					&root->root_key, &root->root_item);
-		if (ret)
-			return ret;
-		spin_lock(&fs_info->fs_roots_lock);
-		btrfs_qgroup_free_meta_all_pertrans(root);
 	}
-	spin_unlock(&fs_info->fs_roots_lock);
+	spin_unlock(&fs_info->fs_roots_radix_lock);
 	return 0;
 }