Re: mmotm 2009-08-24-16-24 uploaded

[Oleg Nesterov <oleg@redhat.com>]

On 08/27, KAMEZAWA Hiroyuki wrote:
>
> On Thu, 27 Aug 2009 11:34:41 +0200
> Oleg Nesterov <oleg@redhat.com> wrote:
>
> > On 08/27, KAMEZAWA Hiroyuki wrote:
> > >
> > > On Thu, 27 Aug 2009 14:44:53 +0900
> > > KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> > >
> > > >
> > > > In the newest mmotom, my S14nfslock hangs up. (x86-64/Fedora10)
> > > >
> > > > On Mon, 24 Aug 2009 16:28:30 -0700
> > > > akpm@linux-foundation.org wrote:
> > > >
> > > > > ptrace-__ptrace_detach-do-__wake_up_parent-if-we-reap-the-tracee.patch
> > > > > do_wait-wakeup-optimization-shift-security_task_wait-from-eligible_child-to-wait_consider_task.patch
> > > >
> > > > bisected. following 2 patches for filtering SIGCHLD cause hang (for my environ).
> > > >
> > > > > do_wait-wakeup-optimization-change-__wake_up_parent-to-use-filtered-wakeup.patch
> > > > > do_wait-wakeup-optimization-change-__wake_up_parent-to-use-filtered-wakeup-selinux_bprm_committed_creds-use-__wake_up_parent.patch
> >
> > Confused. Which patch causes the hang? They should be applied in reverse order,
> >
> > 	do_wait-wakeup-optimization-change-__wake_up_parent-to-use-filtered-wakeup-selinux_bprm_committed_creds-use-__wake_up_parent.patch
> > 	do_wait-wakeup-optimization-change-__wake_up_parent-to-use-filtered-wakeup.patch
> >
> > > removed S14nfslockd from rc5.d and check it by strace
> > > ==
> > >  2712] fstat(6, {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
> > > [pid  2712] mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fc6f263c000
> > > [pid  2712] dup(6)                      = 7
> > > [pid  2712] write(6, "2712\n"..., 5)    = 5
> > > [pid  2712] close(6)                    = 0
> > > [pid  2712] munmap(0x7fc6f263c000, 4096) = 0
> > > [pid  2712] clone(Process 2713 attached
> > > child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, child_tidptr=0x7fc6f2625780) = 2713
> > > [pid  2712] wait4(2713, Process 2712 suspended
> > >  <unfinished ...>
> > > ==
> > > When process 2713 exits, process 2712 don't wake up.
> >
> > Hmm, very strange. How can I reproduce?
> >
> Sorry, I don't know.
>
> But exited process's, but not caught, p->exit_signal was -1. (confirmed by printk)
> (details in another mail)

Ah, I didn't notice "Process 2713 attached" above, I guess you did strace -f.

The child was reaped by strace, because

> Name:   rpc.statd
> State:  S (sleeping)
> ...
> SigIgn: 0000000000011000

indeed, SIGCHLD is ignored.

OK, I seem to understand what happens. Could you try the patch below?

Oleg.

--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -1564,9 +1564,6 @@ static int child_wait_callback(wait_queu
 						child_wait);
 	struct task_struct *p = key;
 
-	if (!eligible_child(wo, p))
-		return 0;
-
 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
 		return 0;