Full_Name: Hallvard B Furuseth Version: HEAD, RE23 OS: URL: Submission from: (NULL) (129.240.202.105) Submitted by: hallvard

As far as I can tell from tpool.c, ldap_pvt_thread_pool_destroy() will act prematurely if another thread pauses: ldap_pvt_thread_pool_destroy (...) { ... if ( pool->ltp_open_count ) { ldap_pvt_thread_cond_broadcast(&pool->ltp_cond); ldap_pvt_thread_cond_wait(&pool->ltp_cond, &pool->ltp_mutex); } ... since ldap_pvt_thread_pool_resume() broadcasts on the same mutex. Also ldap_pvt_thread_pool_resume() will reset ltp_state to RUNNING instead of FINISHING/STOPPING.

If that can happen, the fix would be: (a) use while() instead of if(), (b) keep an ltp_unpaused_state along with ltp_pause which is _pause does not change and _resume copies to ltp_state, or just remove LDAP_INT_THREAD_POOL_PAUSING in favor of an ltp_pausing flag. (And possibly some code which currently checks ltp_state ought to check ltp_unpaused_state instead, I'm not sure.) (c) if _destroy() can be called while LDAP_INT_THREAD_POOL_PAUSING, I think the other if()s around ltp_cond need to be while() too.