[bug report] btrfs: get fs_info from block group in write_pinned_extent_entries

[bug report] btrfs: get fs_info from block group in write_pinned_extent_entries

[Dan Carpenter]

Hello David Sterba,

This is a semi-automatic email about new static checker warnings.

The patch b4c7c5d50791: "btrfs: get fs_info from block group in
write_pinned_extent_entries" from Mar 20, 2019, leads to the
following Smatch complaint:

    fs/btrfs/free-space-cache.c:1052 write_pinned_extent_entries()
    warn: variable dereferenced before check 'block_group' (see line 1047)

fs/btrfs/free-space-cache.c
  1046	{
  1047		struct btrfs_fs_info *fs_info = block_group->fs_info;
                                                ^^^^^^^^^^^^^^^^^^^^
The patch adds a dereference

  1048		u64 start, extent_start, extent_end, len;
  1049		struct extent_io_tree *unpin = NULL;
  1050		int ret;
  1051	
  1052		if (!block_group)
                     ^^^^^^^^^^^
But the old code assumed it could be NULL.

  1053			return 0;
  1054	

See also:
fs/btrfs/free-space-cache.c:1319 __btrfs_write_out_cache() warn: variable dereferenced before check 'block_group' (see line 1295)
fs/btrfs/free-space-cache.c:1295 __btrfs_write_out_cache() error: we previously assumed 'block_group' could be null (see line 1253)

regards,
dan carpenter

Re: [bug report] btrfs: get fs_info from block group in write_pinned_extent_entries

[David Sterba]

On Wed, Apr 24, 2019 at 12:52:54PM +0300, Dan Carpenter wrote:
> Hello David Sterba,
> 
> This is a semi-automatic email about new static checker warnings.
> 
> The patch b4c7c5d50791: "btrfs: get fs_info from block group in
> write_pinned_extent_entries" from Mar 20, 2019, leads to the
> following Smatch complaint:
> 
>     fs/btrfs/free-space-cache.c:1052 write_pinned_extent_entries()
>     warn: variable dereferenced before check 'block_group' (see line 1047)
> 
> fs/btrfs/free-space-cache.c
>   1046	{
>   1047		struct btrfs_fs_info *fs_info = block_group->fs_info;
>                                                 ^^^^^^^^^^^^^^^^^^^^
> The patch adds a dereference
> 
>   1048		u64 start, extent_start, extent_end, len;
>   1049		struct extent_io_tree *unpin = NULL;
>   1050		int ret;
>   1051	
>   1052		if (!block_group)
>                      ^^^^^^^^^^^
> But the old code assumed it could be NULL.

Thanks for the report. The pointer will never be null as the caller will
pass a valid pointer, so the dereference is ok. This extra check is
redundant so I'll remove it.