Btrfs: Make sure page mapping dirty tag is properly cleared

Signed-off-by: Chris Mason <chris.mason@oracle.com>

1 files changed, 20 insertions, 2 deletions

diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 7fd4eb7a8f03..a4e9096754fc 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -1825,12 +1825,18 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
                 if (ret)
                         SetPageError(page);
                 else {
-                        unsigned long nr = end_index + 1;
+                        unsigned long max_nr = end_index + 1;
                         set_range_writeback(tree, cur, cur + iosize - 1);
+                        if (!PageWriteback(page)) {
+                                printk("warning page %lu not writeback, "
+                                       "cur %llu end %llu\n", page->index,
+                                       (unsigned long long)cur,
+                                       (unsigned long long)end);
+                        }
 
                         ret = submit_extent_page(WRITE, tree, page, sector,
                                                  iosize, page_offset, bdev,
-                                                 &epd->bio, nr,
+                                                 &epd->bio, max_nr,
                                                  end_bio_extent_writepage);
                         if (ret)
                                 SetPageError(page);
@@ -1840,6 +1846,11 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
                 nr++;
         }
 done:
+        if (nr == 0) {
+                /* make sure the mapping tag for page dirty gets cleared */
+                set_page_writeback(page);
+                end_page_writeback(page);
+        }
         unlock_extent(tree, start, page_end, GFP_NOFS);
         unlock_page(page);
         return 0;
@@ -2408,6 +2419,13 @@ int clear_extent_buffer_dirty(struct extent_map_tree *tree,
                         }
                 }
                 clear_page_dirty_for_io(page);
+                write_lock_irq(&page->mapping->tree_lock);
+                if (!PageDirty(page)) {
+                        radix_tree_tag_clear(&page->mapping->page_tree,
+                                                page_index(page),
+                                                PAGECACHE_TAG_DIRTY);
+                }
+                write_unlock_irq(&page->mapping->tree_lock);
                 unlock_page(page);
         }
         return 0;