mbox series

[0/6] bpf: task_group_seq_get_next: use __next_thread()

Message ID 20230825161842.GA16750@redhat.com (mailing list archive)
Headers show
Series bpf: task_group_seq_get_next: use __next_thread() | expand

Message

Oleg Nesterov Aug. 25, 2023, 4:18 p.m. UTC
Compile tested, 1-5 need the review from bpf maintainers, quite possibly
I did some silly mistakes. I tried to cleanup this code because I could
not look at it, but it has other problems and imo should be rewritten.

6/6 obviously depends on

	[PATCH 1/2] introduce __next_thread(), fix next_tid() vs exec() race
	https://lore.kernel.org/all/20230824143142.GA31222@redhat.com/

which was not merged yet.

To simplify the review, this is the code after 6/6:

	static struct task_struct *task_group_seq_get_next(struct bpf_iter_seq_task_common *common,
							   u32 *tid,
							   bool skip_if_dup_files)
	{
		struct task_struct *task;
		struct pid *pid;
		u32 next_tid;

		if (!*tid) {
			/* The first time, the iterator calls this function. */
			pid = find_pid_ns(common->pid, common->ns);
			task = get_pid_task(pid, PIDTYPE_TGID);
			if (!task)
				return NULL;

			*tid = common->pid;
			common->pid_visiting = common->pid;

			return task;
		}

		/* If the control returns to user space and comes back to the
		 * kernel again, *tid and common->pid_visiting should be the
		 * same for task_seq_start() to pick up the correct task.
		 */
		if (*tid == common->pid_visiting) {
			pid = find_pid_ns(common->pid_visiting, common->ns);
			task = get_pid_task(pid, PIDTYPE_PID);

			return task;
		}

		task = find_task_by_pid_ns(common->pid_visiting, common->ns);
		if (!task)
			return NULL;

	retry:
		task = __next_thread(task);
		if (!task)
			return NULL;

		next_tid = __task_pid_nr_ns(task, PIDTYPE_PID, common->ns);
		if (!next_tid)
			goto retry;

		if (skip_if_dup_files && task->files == task->group_leader->files)
			goto retry;

		*tid = common->pid_visiting = next_tid;
		get_task_struct(task);
		return task;
	}

Oleg.