Btrfs: Make sure page mapping dirty tag is properly cleared

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

diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 7fd4eb7a8f0371f9345107749bc9380240fcc25a..a4e9096754fc17706db05bc4184bfeee8ad8a40f 100644 (file)
--- 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;