Commit 21faded5 authored by Gopal Shankar's avatar Gopal Shankar

Bug#12636001 : deadlock from thd_security_context

PROBLEM:
Threads end-up in deadlock due to locks acquired as described
below,

con1: Run Query on a table. 
  It is important that this SELECT must back-off while
  trying to open the t1 and enter into wait_for_condition().
  The SELECT then is blocked trying to lock mysys_var->mutex
  which is held by con3. The very significant fact here is
  that mysys_var->current_mutex will still point to LOCK_open,
  even if LOCK_open is no longer held by con1 at this point.

con2: Try dropping table used in con1 or query some table.
  It will hold LOCK_open and be blocked trying to lock
  kernel_mutex held by con4.

con3: Try killing the query run by con1.
  It will hold THD::LOCK_thd_data belonging to con1 while
  trying to lock mysys_var->current_mutex belonging to con1.
  But current_mutex will point to LOCK_open which is held
  by con2.

con4: Get innodb engine status
  It will hold kernel_mutex, trying to lock THD::LOCK_thd_data
  belonging to con1 which is held by con3.

So while technically only con2, con3 and con4 participate in the
deadlock, con1's mysys_var->current_mutex pointing to LOCK_open
is a vital component of the deadlock.

CYCLE = (THD::LOCK_thd_data -> LOCK_open ->
         kernel_mutex -> THD::LOCK_thd_data)

FIX:
LOCK_thd_data has responsibility of protecting,
1) thd->query, thd->query_length
2) VIO
3) thd->mysys_var (used by KILL statement and shutdown)
4) THD during thread delete.

Among above responsibilities, 1), 2)and (3,4) seems to be three
independent group of responsibility. If there is different LOCK
owning responsibility of (3,4), the above mentioned deadlock cycle
can be avoid. This fix introduces LOCK_thd_kill to handle
responsibility (3,4), which eliminates the deadlock issue.

Note: The problem is not found in 5.5. Introduction MDL subsystem 
caused metadata locking responsibility to be moved from TDC/TC to
MDL subsystem. Due to this, responsibility of LOCK_open is reduced. 
As the use of LOCK_open is removed in open_table() and 
mysql_rm_table() the above mentioned CYCLE does not form.
Revision ID for changes,
open_table() = dlenev@mysql.com-20100727133458-m3ua9oslnx8fbbvz
mysql_rm_table() = jon.hauglid@oracle.com-20101116100012-kxep9txz2fxy3nmw
parent 3e1758d3
......@@ -632,13 +632,13 @@ Event_scheduler::stop()
DBUG_PRINT("info", ("Scheduler thread has id %lu",
scheduler_thd->thread_id));
/* Lock from delete */
pthread_mutex_lock(&scheduler_thd->LOCK_thd_data);
pthread_mutex_lock(&scheduler_thd->LOCK_thd_kill);
/* This will wake up the thread if it waits on Queue's conditional */
sql_print_information("Event Scheduler: Killing the scheduler thread, "
"thread id %lu",
scheduler_thd->thread_id);
scheduler_thd->awake(THD::KILL_CONNECTION);
pthread_mutex_unlock(&scheduler_thd->LOCK_thd_data);
pthread_mutex_unlock(&scheduler_thd->LOCK_thd_kill);
/* thd could be 0x0, when shutting down */
sql_print_information("Event Scheduler: "
......
......@@ -515,7 +515,7 @@ terminate_slave_thread(THD *thd,
int error;
DBUG_PRINT("loop", ("killing slave thread"));
pthread_mutex_lock(&thd->LOCK_thd_data);
pthread_mutex_lock(&thd->LOCK_thd_kill);
#ifndef DONT_USE_THR_ALARM
/*
Error codes from pthread_kill are:
......@@ -526,7 +526,7 @@ terminate_slave_thread(THD *thd,
DBUG_ASSERT(err != EINVAL);
#endif
thd->awake(THD::NOT_KILLED);
pthread_mutex_unlock(&thd->LOCK_thd_data);
pthread_mutex_unlock(&thd->LOCK_thd_kill);
/*
There is a small chance that slave thread might miss the first
......
......@@ -2203,6 +2203,8 @@ void wait_for_condition(THD *thd, pthread_mutex_t *mutex, pthread_cond_t *cond)
*/
pthread_mutex_unlock(mutex);
DEBUG_SYNC(thd, "waiting_for_table_unlock");
DBUG_EXECUTE_IF("sleep_after_waiting_for_table", my_sleep(1000000););
pthread_mutex_lock(&thd->mysys_var->mutex);
thd->mysys_var->current_mutex= 0;
thd->mysys_var->current_cond= 0;
......
......@@ -365,6 +365,7 @@ extern "C"
char *thd_security_context(THD *thd, char *buffer, unsigned int length,
unsigned int max_query_len)
{
DEBUG_SYNC(thd, "thd_security_context");
String str(buffer, length, &my_charset_latin1);
const Security_context *sctx= &thd->main_security_ctx;
char header[64];
......@@ -695,6 +696,7 @@ THD::THD()
active_vio = 0;
#endif
pthread_mutex_init(&LOCK_thd_data, MY_MUTEX_INIT_FAST);
pthread_mutex_init(&LOCK_thd_kill, MY_MUTEX_INIT_FAST);
/* Variables with default values */
proc_info="login";
......@@ -999,6 +1001,8 @@ THD::~THD()
/* Ensure that no one is using THD */
pthread_mutex_lock(&LOCK_thd_data);
pthread_mutex_unlock(&LOCK_thd_data);
pthread_mutex_lock(&LOCK_thd_kill);
pthread_mutex_unlock(&LOCK_thd_kill);
add_to_status(&global_status_var, &status_var);
/* Close connection */
......@@ -1026,6 +1030,7 @@ THD::~THD()
#endif
mysys_var=0; // Safety (shouldn't be needed)
pthread_mutex_destroy(&LOCK_thd_data);
pthread_mutex_destroy(&LOCK_thd_kill);
#ifndef DBUG_OFF
dbug_sentry= THD_SENTRY_GONE;
#endif
......@@ -1104,9 +1109,11 @@ void add_diff_to_status(STATUS_VAR *to_var, STATUS_VAR *from_var,
void THD::awake(THD::killed_state state_to_set)
{
DBUG_ENTER("THD::awake");
DBUG_PRINT("enter", ("this: 0x%lx", (long) this));
DBUG_PRINT("enter", ("this: 0x%lx thread_id=%lu killed_state=%d",
(long) this, thread_id, state_to_set));
THD_CHECK_SENTRY(this);
safe_mutex_assert_owner(&LOCK_thd_data);
safe_mutex_assert_not_owner(&LOCK_thd_data);
safe_mutex_assert_owner(&LOCK_thd_kill);
killed= state_to_set;
if (state_to_set != THD::KILL_QUERY)
......@@ -1127,7 +1134,9 @@ void THD::awake(THD::killed_state state_to_set)
hack is not used.
*/
pthread_mutex_lock(&LOCK_thd_data);
close_active_vio();
pthread_mutex_unlock(&LOCK_thd_data);
}
#endif
}
......
......@@ -1342,11 +1342,23 @@ public:
Protects THD data accessed from other threads:
- thd->query and thd->query_length (used by SHOW ENGINE
INNODB STATUS and SHOW PROCESSLIST
- thd->mysys_var (used by KILL statement and shutdown).
Is locked when THD is deleted.
*/
pthread_mutex_t LOCK_thd_data;
/**
- Protects thd->mysys_var (used during KILL statement and shutdown).
- Is Locked when THD is deleted.
Note: This responsibility was earlier handled by LOCK_thd_data.
This lock is introduced to solve a deadlock issue waiting for
LOCK_thd_data. As this lock reduces responsibility of LOCK_thd_data
the deadlock issues is solved.
Caution: LOCK_thd_kill should not be taken while holding LOCK_thd_data.
THD::awake() currently takes LOCK_thd_data after holding
LOCK_thd_kill.
*/
pthread_mutex_t LOCK_thd_kill;
/* all prepared statements and cursors of this connection */
Statement_map stmt_map;
/*
......
......@@ -7187,7 +7187,7 @@ uint kill_one_thread(THD *thd, ulong id, bool only_kill_query)
continue;
if (tmp->thread_id == id)
{
pthread_mutex_lock(&tmp->LOCK_thd_data); // Lock from delete
pthread_mutex_lock(&tmp->LOCK_thd_kill); // Lock from delete
break;
}
}
......@@ -7215,12 +7215,13 @@ uint kill_one_thread(THD *thd, ulong id, bool only_kill_query)
if ((thd->security_ctx->master_access & SUPER_ACL) ||
thd->security_ctx->user_matches(tmp->security_ctx))
{
DEBUG_SYNC(thd, "kill_one_thread_before_kill");
tmp->awake(only_kill_query ? THD::KILL_QUERY : THD::KILL_CONNECTION);
error=0;
}
else
error=ER_KILL_DENIED_ERROR;
pthread_mutex_unlock(&tmp->LOCK_thd_data);
pthread_mutex_unlock(&tmp->LOCK_thd_kill);
}
DBUG_PRINT("exit", ("%d", error));
DBUG_RETURN(error);
......
......@@ -1133,7 +1133,7 @@ void kill_zombie_dump_threads(uint32 slave_server_id)
if (tmp->command == COM_BINLOG_DUMP &&
tmp->server_id == slave_server_id)
{
pthread_mutex_lock(&tmp->LOCK_thd_data); // Lock from delete
pthread_mutex_lock(&tmp->LOCK_thd_kill); // Lock from delete
break;
}
}
......@@ -1146,7 +1146,7 @@ void kill_zombie_dump_threads(uint32 slave_server_id)
again. We just to do kill the thread ourselves.
*/
tmp->awake(THD::KILL_QUERY);
pthread_mutex_unlock(&tmp->LOCK_thd_data);
pthread_mutex_unlock(&tmp->LOCK_thd_kill);
}
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment