[$] LWN.net Weekly Edition for August 6, 2026

A new API proposal aims to streamline the process of creating new Linux processes, offering an alternative to the traditional fork()/exec() method. The concept, presented by Li Chen, involves developing a process-builder API that efficiently assembles new processes from scratch, without duplicating the parent process.

In the current system, the fork() call duplicates a process, which can be inefficient as most of the parent's state is discarded before the execve() function is called to run a new program. Chen's proposed process-builder API seeks to create a new process without relying heavily on the parent's state, enhancing efficiency.

To initiate this process, Chen proposes using the existing pidfd_open() system call, which creates an empty process shell. This shell has no process ID, kernel task structure, or impact on the parent's process-count limit. The shell is then filled with information to build and run the new process, using a structure called struct pidfd_spawn_run_args.

The structure includes fields for the executable path, argument and environment arrays, and flags indicating the type of actions to be performed. The flags field must be zero, and the reserved fields are ignored. The type of actions to be taken before the process is instantiated and launched are specified in the actions field, which points to an array of struct pidfd_spawn_action structures.

These actions include duplicating file descriptors, closing a range of file descriptors, changing the new process's working directory, and adjusting other process settings. By providing this information, the process-builder API enables the creation of new processes more efficiently, without the need for the parent process's state to be duplicated and subsequently discarded.