Re: [PATCH 1/3] btrfs: remove unnecessary -ENOMEM BUG_ON check in extent-tree.c/exclude_super_stripes

[David Sterba <dave@jikos.cz>]

On Thu, Sep 06, 2012 at 02:40:41PM +0800, Wang Sheng-Hui wrote:
> The memory allocation failure is BUG_ON in add_excluded_extent (following
> the code path) and btrfs_rmap_block. No need to BUG_ON -ENOMEM inside
> exclude_super_stripes itself.

No please.

> Its return value is always 0, and useless for its callers. Set it as void
> instead 0-returned.

btrfs_rmap_block itself contains a BUG_ON:

3980 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3981                      u64 chunk_start, u64 physical, u64 devid,
3982                      u64 **logical, int *naddrs, int *stripe_len)
3983 {
3984         struct extent_map_tree *em_tree = &map_tree->map_tree;
3985         struct extent_map *em;
3986         struct map_lookup *map;
3987         u64 *buf;
3988         u64 bytenr;
3989         u64 length;
3990         u64 stripe_nr;
3991         int i, j, nr = 0;
3992
3993         read_lock(&em_tree->lock);
3994         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3995         read_unlock(&em_tree->lock);
3996
3997         BUG_ON(!em || em->start != chunk_start);

And this should be turned into an 'return error', thus giving a non-zero return
code that should be handled in the callers.

Eg. this patch attempts to do that
http://www.mail-archive.com/linux-btrfs@vger.kernel.org/msg15470.html

but has not been merged due to incorrect fix inside exclude_super_stripes
(introduced in the patch).

The same objection for return code cleanups will hold for any function that
returns 0 but is full of BUG_ONs.


david