Re: [PATCH v2] Btrfs: make xattr replace operations atomic

[Robert White <rwhite@pobox.com>]

On 11/07/2014 05:01 PM, Filipe Manana wrote:
> V2: Added missing btrfs_mark_buffer_dirty() call for the case where a replace
>      happens for an item with no other xattrs and the new xattr value size is
>      the same as the old xattr value size;
>      Made btrfs_search_slot not release the path if it returns EOVERFLOW (from
>      split_leaf), so that we can get the current size of the item when doing
>      the replace and extend the leaf to a size smaller then what we passed to
>      btrfs_search_slot if the xattr exists in the item (and the new extend size
>      is new_size - old_size, excluding xattr's name and sizeof struct btrfs_dir_item).

Stupid Question Here: How does this "replace and extend" affect the 
whole copy-on-write paradigm. Aren't you now overwriting the contents of 
the old data instead of doing a COW?

Not a passive-aggressive commentary question, I genuinely don't 
understand this piece of code so I am seeking a learning opportunity.