btrfs: add back missing dirty page rate limiting to defrag

A defrag operation can dirty a lot of pages, specially if operating on
the entire file or a large file range. Any task dirtying pages should
periodically call balance_dirty_pages_ratelimited(), as stated in that
function's comments, otherwise they can leave too many dirty pages in
the system. This is what we did before the refactoring in 5.16, and
it should have remained, just like in the buffered write path and
relocation. So restore that behaviour.

Fixes: 7b50803 ("btrfs: defrag: use defrag_one_cluster() to implement btrfs_defrag_file()")
CC: [email protected] # 5.16
Reviewed-by: Qu Wenruo <[email protected]>
Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: David Sterba <[email protected]>

Author: fdmanana

fs/btrfs/ioctl.c

@@ -1579,6 +1579,7 @@ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 	}
 
 	while (cur < last_byte) {
+		const unsigned long prev_sectors_defragged = sectors_defragged;
 		u64 cluster_end;
 
 		/* The cluster size 256K should always be page aligned */
@@ -1610,6 +1611,10 @@ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 				cluster_end + 1 - cur, extent_thresh,
 				newer_than, do_compress,
 				&sectors_defragged, max_to_defrag);
+
+		if (sectors_defragged > prev_sectors_defragged)
+			balance_dirty_pages_ratelimited(inode->i_mapping);
+
 		btrfs_inode_unlock(inode, 0);
 		if (ret < 0)
 			break;