Re: [RFC PATCH 3/X] ptrace: introduce the empty "struct ptrace_task"

[Roland McGrath <roland@redhat.com>]

>  /**
> + * tracehook_init_task - initialize the new child
> + * @child:		new child task

s/child/task/

> + * @clone_flags:	%CLONE_* flags from clone/fork/vfork system call
> + * @trace:		return value from tracehook_prepare_clone()
> + *
> + * This is called immediately after dup_task_struct().

    * It must clear/reset any tracing state so that tracehook_free_task()
    * will work safely if the task creation fails.  If the task creation 
    * succeeds, a tracehook_finish_clone() call will follow with locks
    * held, before @task starts or is accessible to anyone else.

> +int alloc_ptrace_task(struct task_struct *tsk)

This deserves a short comment about the context it's called from,
and when it is or isn't called at all.

> +	if (cmpxchg(&tsk->ptrace_task, NULL, ptrace_task) != NULL)
> +		kfree(ptrace_task);

I don't see cmpxchg() used very often at all in generic kernel code.  I
wonder how good a choice it is across every arch.  Is there a reason not to
use e.g. task_lock() to mediate installing a new ->ptrace_task pointer?
That seems like a more conservative route.


Thanks,
Roland