Hello,
1. What would be an ideal way to check if a task is alive or dead from
kernel space?
I dont want to check current execution state of the task, rather I
want to check whether task
with this pid is existing currently or not? I might want to do it
periodically, hence even though
a certain pid might exist, it might have got recycled by the time I
check again, right?
Following is an elementary attempt to achieve the same, but I am not
convinced much, and
I am not even sure I can do this kind of thing without knowing the
context I am executing in.
task_t *p;
read_lock(&tasklist_lock);
for_each_process(p) {
if ( strcmp (p->comm, $your-daemon-name) == 0)
break;
}
read_unlock(&tasklist_lock);
2. Basically, what would be parameters of a task which will uniquely
identify a task which ever
existed on my machine. Would it be <name, pid> tuple? Or something
else which uniquely
does this job? Basically differentiate N user space threads in a process.
3. What is the algorithm followed for recycling pids? It does not look
very straight fwd like
pick up the latest freed pid for new task etc. I guess the pid
recycling must be starting after
certain threshold.
Any pointers will be helpful.
-Leo.
Leonidas . wrote:
> 1. What would be an ideal way to check if a task is alive or dead from
> kernel space?
Get a reference to the task's pid (call get_task_pid(), or get_pid() on
the return value of task_pid()), then later check whether pid_task()
works.
(pid_alive() is not what you want because this would require that you
hold a reference to the task_struct.)
> even though a certain pid might exist, it might have got recycled by
> the time I check again, right?
The functions above do not work with PID numbers but with struct pid
which is a reference-counted object. (The functions with "get" in their
name increase the reference count, so don't forget to put_pid() when you
no longer need it.) See also the big comment in include/linux/pid.h.
HTH
Clemens
On Fri, Oct 23, 2009 at 1:43 PM, Clemens Ladisch <[email protected]> wrote:
> Leonidas . wrote:
>> 1. What would be an ideal way to check if a task is alive or dead from
>> kernel space?
>
> Get a reference to the task's pid (call get_task_pid(), or get_pid() on
> the return value of task_pid()), then later check whether pid_task()
> works.
>
> (pid_alive() is not what you want because this would require that you
> hold a reference to the task_struct.)
>
>> even though a certain pid might exist, it might have got recycled by
>> the time I check again, right?
>
> The functions above do not work with PID numbers but with struct pid
> which is a reference-counted object. ?(The functions with "get" in their
> name increase the reference count, so don't forget to put_pid() when you
> no longer need it.) ?See also the big comment in include/linux/pid.h.
>
>
> HTH
> Clemens
>
Yes, the comment in pid.h says it.
Was going through pid.c, what is the fundamental difference between pid_task()
and get_pid_task()? Is it correct to say that, get_pid_task() will
check whether the
task struct is stale or not and return accordingly and pid_task() will
blindly return
task_struct which might be stale?
Now my understanding is get_pid_task() should be followed by put_pid_task()
so the reference counting work as expected, but put_pid_task() is not an
exported symbol? Am I missing here something?
The obvious question which follows from above is what would be the correct
way to determine whether a process is alive or not using pid_alive()? Using
pid_task() does not seem correct and seemingly correct way looks unfeasible.
-Leo.
> The obvious question which follows from above is what would be the correct
> way to determine whether a process is alive or not using pid_alive()? Using
> pid_task() does not seem correct and seemingly correct way looks unfeasible.
>
I guess, just checking task_struct->exit_state will be okay to do,
right? If exit
state is DEAD then the task is stale, might avoid using the reference counting
counterparts, right?
-Leo.
Leonidas . wrote:
> On Fri, Oct 23, 2009 at 1:43 PM, Clemens Ladisch <[email protected]> wrote:
> > Get a reference to the task's pid (...), then later check whether
> > pid_task() works.
>
> Was going through pid.c, what is the fundamental difference between pid_task()
> and get_pid_task()?
As I said, the functions with "get" in their name increase the reference
count. In other words, get_pid_task(p) is implemented as
get_task(pid_task(p)), so, if it succeeded, you would then own a
reference to the task.
> Now my understanding is get_pid_task() should be followed by put_pid_task()
> so the reference counting work as expected, but put_pid_task() is not an
> exported symbol?
You'd have a standard task_struct, so you'd just use put_task(). And if
you don't actually want a reference to the task (because you only want
to check whether the task still exists), just use pid_task() instead.
HTH
Clemens