Memory leak from strdup

[jimmy-mcelwain]

Host: YRC1000 running MotoROS2
Client: PC running Ubuntu 22.04
ROS2 version: Humble
Commits: The commit hashes/version used to build MotoROS2 are available here
DDS: FastDDS

I am a contributor to MotoROS2. Our implementation allows the controller and client to be disconnected/reconnected repeatedly. Whenever I disconnect the client pc from the host, the memory is freed and everything is shut down, and then everything is re-initialized upon connecting again. However, I noticed a memory leak, 24 bytes per disconnect/reconnect cycle. I eventually traced it to its source and found the leak.

rcl/rcl/src/rcl/init.c

Lines 216 to 223 in 2615b0e

	if (context->global_arguments.impl->enclave) {
	context->impl->init_options.impl->rmw_init_options.enclave = rcutils_strdup(
	context->global_arguments.impl->enclave,
	context->impl->allocator);
	} else {
	context->impl->init_options.impl->rmw_init_options.enclave = rcutils_strdup(
	"/", context->impl->allocator);
	}

The memory allocated in these strdup calls is never freed.

I have a branch on our fork of micro-ros here where I added a single line freeing the memory allocated in that strdup, and the memory leak disappeared. There is now no memory lost with a connect/disconnect cycle.

I noticed the problem in Motoman's own fork of micro-ros/rcl, which is somewhat outdated. But looking through the up-to-date upstream repos, the problem is seemingly present in micro-ros/rcl as well as here. Not just in humble, but in newer distributions/rolling as well.

I do not know if the spot that I freed the memory is the proper spot to do so. But if I understand correctly, it has not yet been fixed upstream (here) and the proper spot to free the memory is somewhere in this repo.

[fujitatomoya]

which is somewhat outdated. But looking through the up-to-date upstream repos, the problem is seemingly present in micro-ros/rcl as well as here. Not just in humble, but in newer distributions/rolling as well.

did you confirm the there is a memory leak on ros2/rcl (rolling)?

it looks like it frees the memory here.

rcl/rcl/src/rcl/context.c

Line 135 in 2615b0e

rcl_ret_t init_options_fini_ret = rcl_init_options_fini(&(context->impl->init_options));

rcl/rcl/src/rcl/init_options.c

Line 138 in 2615b0e

rmw_ret_t rmw_ret = rmw_init_options_fini(&(init_options->impl->rmw_init_options));

and i see underlying all rmw tier 1 implementation free the memory for enclave.

[jimmy-mcelwain]

Sorry for the late response, but thank you. I was unaware that it was the responsibility of the rmw to free that memory. I thought that it was the responsibility of rcl. I am using rmw_microxrcedds. I took a look there, and it seems to me like enclave is not freed if RCL_MICROROS_COMPLETE_IMPL is set. I opened up an issue at micro-ROS/rmw_microxrcedds#311.