autowarefoundation/agnocast

get_default_domain_id() (agnocast_cie_thread_configurator/src/util.cpp) parses ROS_DOMAIN_ID with std::stoul, which throws uncaught on an empty (ROS_DOMAIN_ID=) or non-numeric value → crashes the node. It also only checks != nullptr, not empty. size_t get_default_domain_id() { const char * env_value = std::getenv("ROS_DOMAIN_ID"); if (env_value != nullptr) return static_cast<size_t>(std::stoul(env_value)); // throws on "" / non-numeric return 0; } Repro The function compiled standalone: cat > /tmp/repro.cpp <<'EOF' #include <cstdio> #include <cstdlib> #include <string> size_t get_default_domain_id() { const char * env_value = std::getenv("ROS_DOMAIN_ID"); if (env_value != nullptr) return static_cast<size_t>(std::stoul(env_value)); return 0; } int main() { printf("domain=%zu\n", get_default_domain_id()); } EOF g++ -std=c++17 /tmp/repro.cpp -o /tmp/repro ROS_DOMAIN_ID=7 /tmp/repro # -> domain=7 (exit 0) ROS_DOMAIN_ID= /tmp/repro # -> terminate (std::invalid_argu

Background agnocast_internal.h documents the kmod's locking convention ("internal/static helpers are lock-free, caller holds the appropriate lock"), but the convention is enforced only by code review. Internal helpers like agnocast_get_size_{sub,pub}_info_htable (need wrapper->topic_rwsem), find_topic / add_topic / agnocast_find_process_info (need global_htables_rwsem) trust callers silently. Proposal Add lockdep_assert_held*() to those helpers. It's a no-op on production kernels (CONFIG_LOCKDEP off) and prints a WARNING: with backtrace in debug kernels — zero release cost, immediate signal during dev/test. int agnocast_get_size_sub_info_htable(struct topic_wrapper *wrapper) { lockdep_assert_held(&wrapper->topic_rwsem); /* ... existing body ... */ } Scope (Stage 1, single PR, no behaviour change) Annotate the helpers that read/walk shared data structures and rely on the caller holding the relevant rwsem. Suggested starting list: agnocast_get_size_sub_info_htable, agnocast_g

Currently, agnocast_ioctl_get_topic_subscriber_info and agnocast_ioctl_get_topic_publisher_info return -ENOBUFS when the endpoint count exceeds the caller's buffer size. While this can't happen in practice, it would be more useful to return partial results along with metadata indicating truncation. Options Return the actual total count in ret_topic_info_ret_num (even when it exceeds buffer size), so callers can detect truncation by comparing against their buffer size Add a boolean field indicating whether more entries exist beyond what was returned

Increasing MAX_RECEIVE_NUM should improve performance by allowing more messages to be received in a single ioctl call. However, the current implementation copies the entire array from kernel to user space, so we cannot increase MAX_RECEIVE_NUM significantly due to the copy overhead. Modify the implementation to copy only the actually received entries (ret_entry_num elements) instead of the full array. Note: When handling larger arrays, copy_to_user may not work correctly if the data crosses page boundaries. This needs to be considered in the implementation. Related Link #937