MDEV-5479: Prevent mysqld from stealing Unix socket of another active instance

When two MariaDB server instances are configured with the same Unix
socket path but different TCP ports, the second instance starting up
would unconditionally unlink() the first instance's active socket file
in network_init(). This silently broke the first instance's ability to
accept local socket connections, with no error or warning.

Root cause:
The call `(void) unlink(mysqld_unix_port)` in network_init()
(sql/mysqld.cc) removed any existing socket file without checking
whether another running server process was actively listening on it.
Notably, the bind() error handler immediately following the unlink
already prints "Do you already have another server running on
socket: %s ?" -- but this message could never trigger, because the
unconditional unlink() removed the socket before bind() had a
chance to fail.

Fix:
Before unlinking an existing socket file, attempt to connect() to it:
- If connect() succeeds, another server is actively using the socket.
  Print an error message and abort startup via unireg_abort(1).
- If connect() fails (ECONNREFUSED or similar), the socket is stale
  from a previous unclean shutdown. Proceed to unlink() as before.
- If the socket file does not exist at all, skip directly to bind().
- If the file exists but is not a socket (S_ISSOCK check), treat it
  as removable to preserve the previous behavior for non-socket files.

Before this patch:
  $ mysqld --port=13306 --socket=/tmp/mysql.sock   # starts fine
  $ mysqld --port=13307 --socket=/tmp/mysql.sock
  # starts, steals socket
  # First instance can no longer accept socket connections

After this patch:
  $ mysqld --port=13306 --socket=/tmp/mysql.sock   # starts fine
  $ mysqld --port=13307 --socket=/tmp/mysql.sock   # refuses to start:
  # "Another server process is already using the socket file
  #  '/tmp/mysql.sock'. Aborting."
  # First instance continues to operate normally

Test plan:
1. Start first MariaDB instance with
   --socket=/tmp/test.sock --port=13306
2. Verify it is listening: mysql --socket=/tmp/test.sock -e "SELECT 1"
3. Attempt to start second instance:
   --socket=/tmp/test.sock --port=13307
4. Verify second instance aborts with the new error message in error log
5. Verify first instance still accepts connections via the socket
6. Stop the first instance, verify a new instance
   can start (stale socket cleanup still works)

Co-Authored-By: Claude AI <noreply@anthropic.com>

Author: FaramosCZ

sql/mysqld.cc

@@ -2779,7 +2779,35 @@ static void network_init(void)
     else
 #endif
     {
-      (void) unlink(mysqld_unix_port);
+      struct stat statbuf;
+      if (stat(mysqld_unix_port, &statbuf) == 0)
+      {
+        /* Socket file exists - check if another server is using it */
+        if (S_ISSOCK(statbuf.st_mode))
+        {
+          int test_fd= socket(AF_UNIX, SOCK_STREAM, 0);
+          if (test_fd >= 0)
+          {
+            struct sockaddr_un test_addr;
+            bzero((char*) &test_addr, sizeof(test_addr));
+            test_addr.sun_family= AF_UNIX;
+            strmov(test_addr.sun_path, mysqld_unix_port);
+            if (connect(test_fd, (struct sockaddr *) &test_addr,
+                        sizeof(test_addr)) == 0)
+            {
+              /* Socket is active - another server is listening */
+              close(test_fd);
+              sql_print_error("Another server process is already "
+                              "using the socket file '%s'. "
+                              "Aborting.", mysqld_unix_port);
+              unireg_abort(1);
+            }
+            close(test_fd);
+          }
+        }
+        /* Socket file is stale or not a socket - safe to remove */
+        (void) unlink(mysqld_unix_port);
+      }
       port_len= sizeof(UNIXaddr);
     }
     arg= 1;