Properly unlock the socket file lock in ansible-connection (#39223)

Also use a lock file per host, rather than one global file lock.

Commit 9c0275a879 introduced a bug where the lock file was only being
unlocked by the child PID of the resulting fork done in ansible-connection.
This causes delays when a large inventory causes a lot of contention on
that global lock. This patch fixes the problem by ensuring the lock is
released regardless of the fork condition, and also to use a lock file
based on the remote address of the target host, removing the global lock
bottleneck.

Fixes #38892

bin/ansible-connection

@@ -20,6 +20,8 @@ import traceback
 import errno
 import json
 
+from contextlib import contextmanager
+
 from ansible import constants as C
 from ansible.module_utils._text import to_bytes, to_native, to_text
 from ansible.module_utils.six import PY3
@@ -33,6 +35,21 @@ from ansible.utils.display import Display
 from ansible.utils.jsonrpc import JsonRpcServer
 
 
+@contextmanager
+def file_lock(lock_path):
+    """
+    Uses contextmanager to create and release a file lock based on the
+    given path. This allows us to create locks using `with file_lock()`
+    to prevent deadlocks related to failure to unlock properly.
+    """
+
+    lock_fd = os.open(lock_path, os.O_RDWR | os.O_CREAT, 0o600)
+    fcntl.lockf(lock_fd, fcntl.LOCK_EX)
+    yield
+    fcntl.lockf(lock_fd, fcntl.LOCK_UN)
+    os.close(lock_fd)
+
+
 class ConnectionProcess(object):
     '''
     The connection process wraps around a Connection object that manages
@@ -209,13 +226,10 @@ def main():
         tmp_path = unfrackpath(C.PERSISTENT_CONTROL_PATH_DIR)
         makedirs_safe(tmp_path)
 
-        lock_path = unfrackpath("%s/.ansible_pc_lock" % tmp_path)
+        lock_path = unfrackpath("%s/.ansible_pc_lock_%s" % (tmp_path, play_context.remote_addr))
         socket_path = unfrackpath(cp % dict(directory=tmp_path))
 
-        # if the socket file doesn't exist, spin up the daemon process
+        with file_lock(lock_path):
-        lock_fd = os.open(lock_path, os.O_RDWR | os.O_CREAT, 0o600)
-        fcntl.lockf(lock_fd, fcntl.LOCK_EX)
-
             if not os.path.exists(socket_path):
                 messages.append('local domain socket does not exist, starting it')
                 original_path = os.getcwd()
@@ -232,9 +246,6 @@ def main():
                         messages.append(traceback.format_exc())
                         rc = 1
 
-                fcntl.lockf(lock_fd, fcntl.LOCK_UN)
-                os.close(lock_fd)
-
                     if rc == 0:
                         process.run()