Btrfs: clear uptodate flags of pages in sys_array eb

We set uptodate flag to pages in the temporary sys_array eb,
but do not clear the flag after free eb.  As the special
btree inode may still hold a reference on those pages, the
uptodate flag can remain alive in them.

If btrfs_super_chunk_root has been intentionally changed to the
offset of this sys_array eb, reading chunk_root will read content
of sys_array and it will skip our beautiful checks in
btree_readpage_end_io_hook() because of
"pages of eb are uptodate => eb is uptodate"

This adds the 'clear uptodate' part to force it to read from disk.

Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 673c72ab4fbe5087d69844420cc99d6b3b0e2df0..42ccde43053bf2bccd5d114ff90624c5ab064637 100644 (file)
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6631,12 +6631,14 @@ int btrfs_read_sys_array(struct btrfs_root *root)
                sb_array_offset += len;
                cur_offset += len;
        }
+       clear_extent_buffer_uptodate(sb);
        free_extent_buffer_stale(sb);
        return ret;
 
 out_short_read:
        printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
                        len, cur_offset);
+       clear_extent_buffer_uptodate(sb);
        free_extent_buffer_stale(sb);
        return -EIO;
 }