2004-04-17 15:41:36

by Fabiano Ramos

[permalink] [raw]
Subject: get_task_struct()

Dear All,

Inside sys_ptrace, the function get_task_struct is invoked after
retrieving the child's task srtuct pointer. Why is it done? I have
tracked down the code and noticed that it is in fact an increment
on the (page?) counter. Can you help me understand it?

Is it necessary to call free_task_struct whenever its get counterpart
is called?


** arch/i386/kernel/ptrace.c
> read_lock(&tasklist_lock);
> child = find_task_by_pid(pid);
> if (child)
> get_task_struct(child);
> read_unlock(&tasklist_lock);

Thanks in advance,
Fabiano


2004-04-17 16:19:55

by Robert Love

[permalink] [raw]
Subject: Re: get_task_struct()

On Sat, 2004-04-17 at 12:46 -0300, Fabiano Ramos wrote:

> Inside sys_ptrace, the function get_task_struct is invoked after
> retrieving the child's task srtuct pointer. Why is it done? I have
> tracked down the code and noticed that it is in fact an increment
> on the (page?) counter. Can you help me understand it?

Think of get_task_struct(foo) as a reference count on foo's task
structure. So long as the reference count is elevated, foo's task
structure cannot be deallocated.

get_task_struct() is called here because sys_ptrace() needs to ensure
that the child's task structure does not go away out from under this
function, but we do not want to hold the tasklist_lock.

> Is it necessary to call free_task_struct whenever its get counterpart
> is called?

Yes, although it is called "put_task_struct()". It is called at the end
of this function.

Robert Love