MDEV-21910 Deadlock between BF abort and manual KILL commandbb-10.1-MDEV-21910-2

When high priority replication slave applier encounters lock conflict in innodb,
it will force the conflicting lock holder transaction (victim) to rollback.
This is a must in multi-master sychronous replication model to avoid cluster lock-up.
This high priority victim abort (aka "brute force" (BF) abort), is started
from innodb lock manager while holding the global lock_sys->mutex and victim's
transaction's (trx) mutex.

Depending on the execution state of the victim transaction, it may happen that the
BF abort will call for THD::awake() to wake up the victim transaction for the rollback.
Now, if BF abort requires THD::awake() to be called, then the applier thread executed
locking protocol of: lock_sys->mutex -> victim THD::LOCK_thd_data

If, at the same time another DBMS super user issues KILL command to abort the same victim,
it will execute locking protocol of: victim THD::LOCK_thd_data  -> lock_sys->mutex.
These two locking protocol acquire mutexes in opposite order, hence unresolvable mutex locking
deadlock may occur.

The fix in this commit adds THD::wsrep_killed flag to synchronize who can kill the victim
This flag is set both when BF is called for from innodb and by KILL command.
Either path of victim killing will bail out if victim's wsrep_killed is already
set to avoid mutex conflicts with the other aborter execution.

A new test case was added in galera.galera_bf_kill.test for scenario where
wsrep applier thread and manual KILL command try to kill same idle victim

1 files changed, 6 insertions, 2 deletions

diff --git a/include/mysql/service_wsrep.h b/include/mysql/service_wsrep.h
index 499fbf2c173..7b943dedff8 100644
--- a/include/mysql/service_wsrep.h
+++ b/include/mysql/service_wsrep.h
@@ -99,7 +99,7 @@ extern struct wsrep_service_st {
   enum wsrep_conflict_state   (*wsrep_thd_get_conflict_state_func)(MYSQL_THD);
   my_bool                     (*wsrep_thd_is_BF_func)(MYSQL_THD , my_bool);
   my_bool                     (*wsrep_thd_is_wsrep_func)(MYSQL_THD thd);
-  char *                      (*wsrep_thd_query_func)(THD *thd);
+  const char *                (*wsrep_thd_query_func)(THD *thd);
   enum wsrep_query_state      (*wsrep_thd_query_state_func)(THD *thd);
   const char *                (*wsrep_thd_query_state_str_func)(THD *thd);
   int                         (*wsrep_thd_retry_counter_func)(THD *thd);
@@ -112,6 +112,7 @@ extern struct wsrep_service_st {
   int                         (*wsrep_trx_order_before_func)(MYSQL_THD, MYSQL_THD);
   void                        (*wsrep_unlock_rollback_func)();
   void                        (*wsrep_set_data_home_dir_func)(const char *data_dir);
+  bool                        (*wsrep_thd_set_wsrep_killed_func)(MYSQL_THD thd);
 } *wsrep_service;
 
 #ifdef MYSQL_DYNAMIC_PLUGIN
@@ -163,6 +164,7 @@ extern struct wsrep_service_st {
 #define wsrep_drupal_282555_workaround get_wsrep_drupal_282555_workaround()
 #define wsrep_recovery get_wsrep_recovery()
 #define wsrep_protocol_version get_wsrep_protocol_version()
+#define wsrep_thd_set_wsrep_killed(T) wsrep_service->wsrep_thd_set_wsrep_killed_func(T)
 
 #else
 
@@ -176,7 +178,7 @@ extern long wsrep_protocol_version;
 
 bool wsrep_consistency_check(THD *thd);
 bool wsrep_prepare_key(const unsigned char* cache_key, size_t cache_key_len, const unsigned char* row_id, size_t row_id_len, struct wsrep_buf* key, size_t* key_len);
-char *wsrep_thd_query(THD *thd);
+const char *wsrep_thd_query(THD *thd);
 const char *wsrep_thd_conflict_state_str(THD *thd);
 const char *wsrep_thd_exec_mode_str(THD *thd);
 const char *wsrep_thd_query_state_str(THD *thd);
@@ -214,8 +216,10 @@ void wsrep_thd_set_conflict_state(THD *thd, enum wsrep_conflict_state state);
 bool wsrep_thd_ignore_table(THD *thd);
 void wsrep_unlock_rollback();
 void wsrep_set_data_home_dir(const char *data_dir);
+bool wsrep_thd_set_wsrep_killed(MYSQL_THD thd);
 
 #endif
+/* set wsrep_killed flag for the target THD */
 
 #ifdef __cplusplus
 }