f2fs: avoid node page to be written twice in gc_node_segment
authorHuang Ying <ying.huang@intel.com>
Sun, 7 Sep 2014 03:05:20 +0000 (11:05 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Tue, 9 Sep 2014 20:15:07 +0000 (13:15 -0700)
In gc_node_segment, if node page gc is run concurrently with node page
writeback, and check_valid_map and get_node_page run after page locked
and before cur_valid_map is updated as below, it is possible for the
page to be written twice unnecessarily.

sync_node_pages
  try_lock_page
  ...
check_valid_map   f2fs_write_node_page
    ...
    write_node_page
      do_write_page
        allocate_data_block
  ...
  refresh_sit_entry /* update cur_valid_map */
  ...
    ...
    unlock_page
get_node_page
...
set_page_dirty
...
f2fs_put_page
  unlock_page

This can be solved via calling check_valid_map after get_node_page again.

Signed-off-by: Huang, Ying <ying.huang@intel.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/gc.c

index 943a31db7cc3679c117716b220a58cce1b50a12d..075ea1eb8fa0bad28317c103ae2bbbbad7db0dfa 100644 (file)
@@ -423,6 +423,12 @@ next_step:
                if (IS_ERR(node_page))
                        continue;
 
+               /* block may become invalid during get_node_page */
+               if (check_valid_map(sbi, segno, off) == 0) {
+                       f2fs_put_page(node_page, 1);
+                       continue;
+               }
+
                /* set page dirty and write it */
                if (gc_type == FG_GC) {
                        f2fs_wait_on_page_writeback(node_page, NODE);