On Thu, Oct 14, 2010 at 12:34:25AM +0200, Clemens Buchacher wrote:
> 
> If lstat returns with an error other than NOENT, or if
> check_ok_to_remove is called with anything other than a directory
> and cache_entry is NULL, we get a segmentation fault. Before, an
> error was simply ignored. I don't know which is worse.

I suppose we only need the following additional changes.

- die if lstat returns an error other than ENOENT.

- Rewrite verify_clean_subdirectory to not require a cache_entry.

- Expose lstat result and path cache to the caller of
  lstat_cache_matchlen() in verify_absent_1().

- rewrite check_leading_path (or 'verify_clean_path') to
  check the full path and return zero if
  + the leading path contains a symlink or
  + the leading path exists, but the full path does not
  and returns the path length of the offending entry otherwise.
  
I think that's manageable and much cleaner. I will start to work on
it as soon as possible.

Thank you for your review.

Clemens