Fix for BUG#2921 "Replication problem on mutex lock in mySQL-4.0.18":

re-using unused LOCK_active_mi to serialize all administrative
commands related to replication:
START SLAVE, STOP SLAVE, RESET SLAVE, CHANGE MASTER, init_slave()
(replication autostart at server startup), end_slave() (replication
autostop at server shutdown), LOAD DATA FROM MASTER.
This protects us against a handful of deadlocks (like BUG#2921
when two START SLAVE, but when two STOP SLAVE too).
Removing unused variables.


sql/item_func.cc:
  We don't need LOCK_active_mi just to MASTER_POS_WAIT().
sql/repl_failsafe.cc:
  no need for macro
sql/set_var.cc:
  no need for macro
sql/slave.cc:
  Re-using unused LOCK_active_mi to serialize all administrative
  commands related to replication:
  START SLAVE, STOP SLAVE, RESET SLAVE, CHANGE MASTER, init_slave()
  (replication autostart at server startup), end_slave() (replication
  autostop at server shutdown), LOAD DATA FROM MASTER.
  This protects us against a handful of deadlocks.
  Removing unused variables.
sql/slave.h:
  Re-using LOCK_active_mi to serialize administrative replication commands.
  Macros unneeded. Removing unneeded variables.
sql/sql_parse.cc:
  found unused variable.
  Replacing macros.
sql/sql_show.cc:
  replacing macros

1 files changed, 17 insertions, 4 deletions

diff --git a/sql/slave.cc b/sql/slave.cc
index 84f4ff4f3e0..30052e874b3 100644
--- a/sql/slave.cc
+++ b/sql/slave.cc
@@ -34,7 +34,6 @@ typedef bool (*CHECK_KILLED_FUNC)(THD*,void*);
 volatile bool slave_sql_running = 0, slave_io_running = 0;
 char* slave_load_tmpdir = 0;
 MASTER_INFO *active_mi;
-volatile int active_mi_in_use = 0;
 HASH replicate_do_table, replicate_ignore_table;
 DYNAMIC_ARRAY replicate_wild_do_table, replicate_wild_ignore_table;
 bool do_table_inited = 0, ignore_table_inited = 0;
@@ -114,8 +113,12 @@ int init_slave()
 {
   DBUG_ENTER("init_slave");
 
-  /* This is called when mysqld starts */
-
+  /*
+    This is called when mysqld starts. Before client connections are
+    accepted. However bootstrap may conflict with us if it does START SLAVE.
+    So it's safer to take the lock.
+  */
+  pthread_mutex_lock(&LOCK_active_mi);
   /*
     TODO: re-write this to interate through the list of files
     for multi-master
@@ -160,9 +163,11 @@ int init_slave()
       goto err;
     }
   }
+  pthread_mutex_unlock(&LOCK_active_mi);
   DBUG_RETURN(0);
 
 err:
+  pthread_mutex_unlock(&LOCK_active_mi);
   DBUG_RETURN(1);
 }
 
@@ -806,7 +811,14 @@ static int end_slave_on_walk(MASTER_INFO* mi, gptr /*unused*/)
 
 void end_slave()
 {
-  /* This is called when the server terminates, in close_connections(). */
+  /*
+    This is called when the server terminates, in close_connections().
+    It terminates slave threads. However, some CHANGE MASTER etc may still be
+    running presently. If a START SLAVE was in progress, the mutex lock below
+    will make us wait until slave threads have started, and START SLAVE
+    returns, then we terminate them here.
+  */
+  pthread_mutex_lock(&LOCK_active_mi);
   if (active_mi)
   {
     /*
@@ -827,6 +839,7 @@ void end_slave()
     delete active_mi;
     active_mi= 0;
   }
+  pthread_mutex_unlock(&LOCK_active_mi);
 }