Btrfs: split up __extent_writepage to lower stack usage

Hello Chris Mason,

The patch 263524b4ac6b: "Btrfs: split up __extent_writepage to lower stack usage" from May 21, 2014, leads to the following static checker warning:

	fs/btrfs/extent_io.c:4071 try_release_extent_state()
	warn: use 'mask' here instead of GFP_XXX?

fs/btrfs/extent_io.c
4053 static int try_release_extent_state(struct extent_map_tree *map,
4054 struct extent_io_tree *tree,
4055 struct page *page, gfp_t mask)
4056 {

  4057          u64 start = page_offset(page);
  4058          u64 end = start + PAGE_CACHE_SIZE - 1;
  4059          int ret = 1;
  4060  
  4061          if (test_range_bit(tree, start, end,
  4062                             EXTENT_IOBITS, 0, NULL))
  4063                  ret = 0;
  4064          else {
  4065                  if ((mask & GFP_NOFS) == GFP_NOFS)
  4066                          mask = GFP_NOFS;
  4067                  /*
  4068                   * at this point we can safely clear everything except the
  4069                   * locked bit and the nodatasum bit
  4070                   */
  4071                  ret = clear_extent_bit(tree, start, end,
  4072                                   ~(EXTENT_LOCKED | EXTENT_NODATASUM),
  4073                                   0, 0, NULL, GFP_ATOMIC);
                                                     ^^^^^^^^^^

It upsets the static checkers to keep "mask" around when we don't use it anymore.

4074
4075 /* if clear_extent_bit failed for enomem reasons,
4076 * we can't allow the release to continue.
4077 */
4078 if (ret < 0)
4079 ret = 0;
4080 else
4081 ret = 1;
4082 }
4083 return ret;
4084 }

regards,
dan carpenter

Thanks Dan, I'll switch this around to make it more clear.

-chris