MDEV-38126 Dead code, race conditions and contentious lock_operations in audit plugin

[dr-m]

• The Jira issue number for this PR is: MDEV-38126

Description

The implementation of the audit plugin apparently follows obsolete 1990 version of ISO/IEC 9899, a.k.a. C90. Atomic memory operations were not introduced to the language until 2011. Therefore, a combination of mutexes and volatile variables is being used. This is prone to race conditions that could explain some failures such as MDEV-34074, because no well-defined consistency model is being followed. Elsewhere in the code base, we are making use of features that were introduced in ISO/IEC 14882:2011, such as std::atomic.

Release Notes

The performance of the server_audit plugin was improved and performance_schema instrumentation for SERVER_AUDIT_plugin::lock_operations removed.

How can this PR be tested?

HammerDB with auditing enabled

Basing the PR against the correct MariaDB version

• This is a new feature or a refactoring, and the PR is based against the main branch.
• This is a bug fix, and the PR is based against the earliest maintained branch in which the bug can be reproduced.

This depends on some write_log() code removal in 7251cbc. It should be technically possible to port some of this to an earlier major version (I tried it on 10.11), but that would have to be tested and reviewed separately.

PR quality check

• I checked the CODING_STANDARDS.md file and my PR conforms to this where appropriate.
• For any trivial modifications to the PR, I am ok with the reviewer making the changes themselves.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[dr-m]

I spotted some more errors:

/home/buildbot/src/plugin/server_audit/server_audit.cc:2337:3: runtime error: call to function logger_init_mutexes through pointer to incorrect function type 'void (*)()'
SUMMARY: UndefinedBehaviorSanitizer: function-type-mismatch /home/buildbot/src/plugin/server_audit/server_audit.cc:2337:3 
/home/buildbot/src/plugin/server_audit/server_audit.cc:2337:3: runtime error: call to function logger_init_mutexes through pointer to incorrect function type 'void (*)()'
SUMMARY: UndefinedBehaviorSanitizer: function-type-mismatch /home/buildbot/src/plugin/server_audit/server_audit.cc:2337:3 

It looks like MDEV-34348 and clang -Wcast-function-type-strict were not implemented for this plugin.

[dr-m]

I could not figure out a solution to the logger_init_mutexes() problem. The following did not work out:

diff --git a/include/mysql/service_logger.h b/include/mysql/service_logger.h
index d11dd4291ae..aa6c34fd755 100644
--- a/include/mysql/service_logger.h
+++ b/include/mysql/service_logger.h
@@ -61,7 +61,7 @@ extern "C" {
 typedef struct logger_handle_st LOGGER_HANDLE;
 
 extern struct logger_service_st {
-  void (*logger_init_mutexes)();
+  void (*logger_init_mutexes)(void);
   LOGGER_HANDLE* (*open)(const char *path,
                          unsigned long long size_limit,
                          unsigned int rotations, size_t buffer_size);
diff --git a/plugin/server_audit/server_audit.cc b/plugin/server_audit/server_audit.cc
index b1cc1c08ce2..5972b09319e 100644
--- a/plugin/server_audit/server_audit.cc
+++ b/plugin/server_audit/server_audit.cc
@@ -74,11 +74,13 @@ static void closelog() {}
 #include <my_global.h>
 #include <my_base.h>
 #include <typelib.h>
+#include <string.h>
+extern "C" {
 #include <mysql/plugin.h>
 #include <mysql/plugin_audit.h>
-#include <string.h>
 #include <mysql/service_logger.h>
 #include "../../mysys/mysys_priv.h"
+}
 #ifndef RTLD_DEFAULT
 #define RTLD_DEFAULT NULL
 #endif

In any case, I think that the logger interface will need to be replaced with something similar to the InnoDB log_sys.buf and log_sys.flush_buf as well as #3925 so that we can allow multiple concurrent writes to the memory buffer. This could significantly improve multi-threaded performance.

[dr-m]

I hope that 5676ecf will not introduce function pointer type mismatch elsewhere.