Re: Linux 2.6.38

[Oleg Nesterov <oleg@redhat.com>]

On 03/14, David Rientjes wrote:
>
> He also presented a test case on linux-mm that caused the oom killer to 
> avoid acting if a thread is ptracing a thread in the exit path with 
> PTRACE_O_TRACEEXIT.  That should be fixed with
>
> 	http://marc.info/?l=linux-mm&m=129997893430351

I don't think it can fix this. I didn't verify this, but the slightly
different test-case below should have the same effect.

But this doesn't matter. We can fix this particular case, and we have
the problems with the coredump anyway.

What I can't understand is what exactly the first patch tries to fix.
When I ask you, you tell me that for_each_process() can miss the group
leader because it can exit before sub-threads. This must not happen,
or we have some serious bug triggered by your workload.

So, once again. Could you please explain the original problem and how
this patch helps?

Oleg.

#include <unistd.h>
#include <signal.h>
#include <pthread.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <assert.h>
#include <stdio.h>

void *tfunc(void* arg)
{
	if (arg) {
		ptrace(PTRACE_TRACEME, 0,0,0);
		raise(SIGSTOP);
		pthread_kill(*(pthread_t*)arg, SIGQUIT);
	}
	pause();
}

int main(void)
{
	int pid;

	if (!fork()) {
		pthread_t thread1, thread2;
		pthread_create(&thread1, NULL, tfunc, NULL);
		pthread_create(&thread2, NULL, tfunc, &thread1);
		pause();
		return 0;
	}

	assert((pid = waitpid(-1, NULL, __WALL)) > 0);
	assert(ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_TRACEEXIT) == 0);
	assert(ptrace(PTRACE_CONT, pid, 0, 0) == 0);
	wait(NULL);

	pause();
	return 0;
}