btrfs: populate fully_remapped_bgs_list on mount

Add a function btrfs_populate_fully_remapped_bgs_list() which gets
called on mount, which looks for fully remapped block groups
(i.e. identity_remap_count == 0) which haven't yet had their chunk
stripes and device extents removed.

This happens when a filesystem is unmounted while async discard has not
yet finished, as otherwise the data range occupied by the chunk stripes
would be permanently unusable.

Reviewed-by: Boris Burkov <boris@bur.io>
Signed-off-by: Mark Harmstone <mark@harmstone.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/block-group.c

@@ -4794,6 +4794,10 @@ void btrfs_mark_bg_fully_remapped(struct btrfs_block_group *bg,
 
 
 	if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
+		spin_lock(&bg->lock);
+		set_bit(BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING, &bg->runtime_flags);
+		spin_unlock(&bg->lock);
+
 		btrfs_discard_queue_work(&fs_info->discard_ctl, bg);
 	} else {
 		spin_lock(&fs_info->unused_bgs_lock);
@@ -4811,3 +4815,74 @@ void btrfs_mark_bg_fully_remapped(struct btrfs_block_group *bg,
 		spin_unlock(&fs_info->unused_bgs_lock);
 	}
 }
+
+/*
+ * Compare the block group and chunk trees, and find any fully-remapped block
+ * groups which haven't yet had their chunk stripes and device extents removed,
+ * and put them on the fully_remapped_bgs list so this gets done.
+ *
+ * This happens when a block group becomes fully remapped, i.e. its last
+ * identity mapping is removed, and the volume is unmounted before async
+ * discard has finished. It's important this gets done as until it is the
+ * chunk's stripes are dead space.
+ */
+int btrfs_populate_fully_remapped_bgs_list(struct btrfs_fs_info *fs_info)
+{
+	struct rb_node *node_bg, *node_chunk;
+
+	node_bg = rb_first_cached(&fs_info->block_group_cache_tree);
+	node_chunk = rb_first_cached(&fs_info->mapping_tree);
+
+	while (node_bg && node_chunk) {
+		struct btrfs_block_group *bg;
+		struct btrfs_chunk_map *map;
+
+		bg = rb_entry(node_bg, struct btrfs_block_group, cache_node);
+		map = rb_entry(node_chunk, struct btrfs_chunk_map, rb_node);
+
+		ASSERT(bg->start == map->start);
+
+		if (!(bg->flags & BTRFS_BLOCK_GROUP_REMAPPED))
+			goto next;
+
+		if (bg->identity_remap_count != 0)
+			goto next;
+
+		if (map->num_stripes == 0)
+			goto next;
+
+		spin_lock(&fs_info->unused_bgs_lock);
+
+		if (list_empty(&bg->bg_list)) {
+			btrfs_get_block_group(bg);
+			list_add_tail(&bg->bg_list, &fs_info->fully_remapped_bgs);
+		} else {
+			list_move_tail(&bg->bg_list, &fs_info->fully_remapped_bgs);
+		}
+
+		spin_unlock(&fs_info->unused_bgs_lock);
+
+		/*
+		 * Ideally we'd want to call btrfs_discard_queue_work() here,
+		 * but it'd do nothing as the discard worker hasn't been
+		 * started yet.
+		 *
+		 * The block group will get added to the discard list when
+		 * btrfs_handle_fully_remapped_bgs() gets called, when we
+		 * commit the first transaction.
+		 */
+		if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
+			spin_lock(&bg->lock);
+			set_bit(BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING, &bg->runtime_flags);
+			spin_unlock(&bg->lock);
+		}
+
+next:
+		node_bg = rb_next(node_bg);
+		node_chunk = rb_next(node_chunk);
+	}
+
+	ASSERT(!node_bg && !node_chunk);
+
+	return 0;
+}

fs/btrfs/block-group.h

@@ -94,6 +94,7 @@ enum btrfs_block_group_flags {
 	 */
 	BLOCK_GROUP_FLAG_NEW,
 	BLOCK_GROUP_FLAG_FULLY_REMAPPED,
+	BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING,
 };
 
 enum btrfs_caching_type {
@@ -418,5 +419,6 @@ int btrfs_use_block_group_size_class(struct btrfs_block_group *bg,
 bool btrfs_block_group_should_use_size_class(const struct btrfs_block_group *bg);
 void btrfs_mark_bg_fully_remapped(struct btrfs_block_group *bg,
 				  struct btrfs_trans_handle *trans);
+int btrfs_populate_fully_remapped_bgs_list(struct btrfs_fs_info *fs_info);
 
 #endif /* BTRFS_BLOCK_GROUP_H */

fs/btrfs/disk-io.c

@@ -3601,6 +3601,14 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
 		goto fail_sysfs;
 	}
 
+	if (btrfs_fs_incompat(fs_info, REMAP_TREE)) {
+		ret = btrfs_populate_fully_remapped_bgs_list(fs_info);
+		if (ret) {
+			btrfs_err(fs_info, "failed to populate fully_remapped_bgs list: %d", ret);
+			goto fail_sysfs;
+		}
+	}
+
 	btrfs_zoned_reserve_data_reloc_bg(fs_info);
 	btrfs_free_zone_cache(fs_info);
 

fs/btrfs/free-space-cache.c

@@ -3068,6 +3068,7 @@ bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
 	bool ret = true;
 
 	if (block_group->flags & BTRFS_BLOCK_GROUP_REMAPPED &&
+	    !test_bit(BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING, &block_group->runtime_flags) &&
 	    block_group->identity_remap_count == 0) {
 		return true;
 	}
@@ -3849,6 +3850,23 @@ void btrfs_trim_fully_remapped_block_group(struct btrfs_block_group *bg)
 	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
 	u64 end = btrfs_block_group_end(bg);
 
+	if (!test_bit(BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING, &bg->runtime_flags)) {
+		bg->discard_cursor = end;
+
+		if (bg->used == 0) {
+			spin_lock(&fs_info->unused_bgs_lock);
+			if (!list_empty(&bg->bg_list)) {
+				list_del_init(&bg->bg_list);
+				btrfs_put_block_group(bg);
+			}
+			spin_unlock(&fs_info->unused_bgs_lock);
+
+			btrfs_mark_bg_unused(bg);
+		}
+
+		return;
+	}
+
 	bytes = end - bg->discard_cursor;
 
 	if (max_discard_size &&

fs/btrfs/relocation.c

@@ -4743,6 +4743,10 @@ int btrfs_last_identity_remap_gone(struct btrfs_chunk_map *chunk_map,
 
 	btrfs_remove_bg_from_sinfo(bg);
 
+	spin_lock(&bg->lock);
+	clear_bit(BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING, &bg->runtime_flags);
+	spin_unlock(&bg->lock);
+
 	ret = remove_chunk_stripes(trans, chunk_map, path);
 	if (unlikely(ret)) {
 		btrfs_abort_transaction(trans, ret);