Fix for:

Bug #4810 "deadlock with KILL when the victim was in a wait state"
(I included mutex unlock into exit_cond() for future safety)
and BUG#4827 "KILL while START SLAVE may lead to replication slave crash"


sql/lock.cc:
  we did exit_cond() before unlock(LOCK_open), which led to deadlocks with THD::awake(). Fixing this.
sql/log.cc:
  mutex unlock is now included in exit_cond()
sql/repl_failsafe.cc:
  we did exit_cond() before unlock(LOCK_rpl_status), which led to deadlocks with THD::awake(). Fixing this.
sql/slave.cc:
  we did exit_cond() before unlock(cond_lock), which led to deadlocks with THD::awake(). Fixing this.
  Fixing also that if killed while waiting for slave thread to start, we don't release the mutex
  (that caused a double release of the mutex => crash).
sql/sql_class.h:
  comments about exit_cond()/enter_cond().
  Mutex unlock is now included in exit_cond() so that it's always done in the good order.
sql/sql_table.cc:
  unlock is now included in exit_cond().

1 files changed, 2 insertions, 7 deletions

diff --git a/sql/log.cc b/sql/log.cc
index a0e2196cc59..e031656cc6e 100644
--- a/sql/log.cc
+++ b/sql/log.cc
@@ -1533,12 +1533,8 @@ bool MYSQL_LOG::write(THD *thd,const char *query, uint query_length,
 
   NOTES
     One must have a lock on LOCK_log before calling this function.
-    This lock will be freed before return!
-
-    The reason for the above is that for enter_cond() / exit_cond() to
-    work the mutex must be got before enter_cond() but releases before
-    exit_cond().
-    If you don't do it this way, you will get a deadlock in THD::awake()
+    This lock will be freed before return! That's required by
+    THD::enter_cond() (see NOTES in sql_class.h).
 */
 
 
@@ -1551,7 +1547,6 @@ the I/O slave thread to update it" :
                                         "Has sent all binlog to slave; \
 waiting for binlog to be updated"); 
   pthread_cond_wait(&update_cond, &LOCK_log);
-  pthread_mutex_unlock(&LOCK_log);		// See NOTES
   thd->exit_cond(old_msg);
 }