Re: back-bdb cache wrong lock ordering

[Howard Chu <hyc@symas.com>]

Yin Wang wrote:
> Our study shows another potential deadlock in back-bdb/cache.c
> OpenLDAP 2.4.8

2.4.8 is a few months old already; 2.4.9 is just about to be released. Your 
study should be working with CVS HEAD, to avoid reporting on bugs that have 
already been fixed. Working with old code isn't useful.

> In bdb_cache_release_all(), thread wants to lock the c_lru_mutex
> at cache.c:1366, while it has the c_rwlock from cache.c:1364

The locks in that function are superfluous, since it can only be called while 
the database is shutting down. There are no other active threads when this 
function runs.

> In bdb_cache_delete_internal(), thread wants to lock the c_rwlock
> at cache.c:1317, while it has the c_lru_mutex from callers,
> bdb_cache_delete() at cache.c:1259, or bdb_cache_lru_purge() at
> cache.c:652.
>
> Note that two deadlock bugs of the same two mutexes have been
> reported and verified in the issue tracking system
>
> http://www.openldap.org/its/index.cgi/Software%20Bugs?id=4254
>
> http://www.openldap.org/its/index.cgi/Software%20Bugs?id=3494

Those bug reports were closed 2-3 years ago. You need to focus on something 
recent in order to have any relevance.

> Yin Wang

--
  -- Howard Chu
  CTO, Symas Corp.           http://www.symas.com
  Director, Highland Sun     http://highlandsun.com/hyc/
  Chief Architect, OpenLDAP  http://www.openldap.org/project/