tracehook breaks dhcpcd child processing

tracehook breaks dhcpcd child processing

[Andi Kleen]

[Rafael, one entry for the regression list]

Roland,

The opensuse 11.0 dhcpcd keeps accumulating zombies of its child scripts
with 2.6.27-rc*. 2.6.26 didn't have this problem. After one night I had more 
zombies than normal processes.

I bisected it down to

commit 2b2a1ff64afbadac842bbc58c5166962cf4f7664
Author: Roland McGrath <roland@redhat.com>
Date:   Fri Jul 25 19:45:54 2008 -0700

    tracehook: death

causing the problem. Please fix.

-Andi

Re: tracehook breaks dhcpcd child processing

[Roland McGrath]

Thanks for the report.

Was this problem after the fixes in:
commit 5861bbfcc10fc0358abf52c7d22850c8d180f0b0
commit 5c7edcd7ee6b77b88252fe4096dce1a46a60c829
?

Do you know the process-level details that cause the problem?
(i.e. exits with what parent/signals/etc conditions)
It may take me a bit to set up an opensuse 11.0 vm to try its dhcpcd.


Thanks,
Roland

Re: tracehook breaks dhcpcd child processing

[Andi Kleen]

On Tue, Aug 19, 2008 at 12:43:18PM -0700, Roland McGrath wrote:
> Thanks for the report.
> 
> Was this problem after the fixes in:
> commit 5861bbfcc10fc0358abf52c7d22850c8d180f0b0
> commit 5c7edcd7ee6b77b88252fe4096dce1a46a60c829
> ?

Yes. I saw it with yesterday's Linus HEAD

> 
> Do you know the process-level details that cause the problem?
> (i.e. exits with what parent/signals/etc conditions)
> It may take me a bit to set up an opensuse 11.0 vm to try its dhcpcd.

Not currently. I'll send you a strace log later.

-Andi

Re: tracehook breaks dhcpcd child processing

[Roland McGrath]

I reproduced it, SA_NOCLDWAIT broke.  I'll have a fix momentarily.

Thanks,
Roland

[PATCH] tracehook: fix SA_NOCLDWAIT

[Roland McGrath]

The following changes since commit 1fca25427482387689fa27594c992a961d98768f:
  Linus Torvalds (1):
        Merge branch 'release' of git://git.kernel.org/.../aegl/linux-2.6

are available in the git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/frob/linux-2.6-utrace.git tracehook

Roland McGrath (1):
      tracehook: fix SA_NOCLDWAIT

 kernel/signal.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)


Thanks,
Roland
---
[PATCH] tracehook: fix SA_NOCLDWAIT

I outwitted myself again in commit 2b2a1ff64afbadac842bbc58c5166962cf4f7664,
and broke the SA_NOCLDWAIT behavior so it leaks zombies.  This fixes it.

Reported-by: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Roland McGrath <roland@redhat.com>
---
 kernel/signal.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/signal.c b/kernel/signal.c
index c539f60..e661b01 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -1338,6 +1338,7 @@ int do_notify_parent(struct task_struct *tsk, int sig)
 	struct siginfo info;
 	unsigned long flags;
 	struct sighand_struct *psig;
+	int ret = sig;
 
 	BUG_ON(sig == -1);
 
@@ -1402,7 +1403,7 @@ int do_notify_parent(struct task_struct *tsk, int sig)
 		 * is implementation-defined: we do (if you don't want
 		 * it, just use SIG_IGN instead).
 		 */
-		tsk->exit_signal = -1;
+		ret = tsk->exit_signal = -1;
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = -1;
 	}
@@ -1411,7 +1412,7 @@ int do_notify_parent(struct task_struct *tsk, int sig)
 	__wake_up_parent(tsk, tsk->parent);
 	spin_unlock_irqrestore(&psig->siglock, flags);
 
-	return sig;
+	return ret;
 }
 
 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)

Re: [PATCH] tracehook: fix SA_NOCLDWAIT

[Andi Kleen]

> I outwitted myself again in commit 2b2a1ff64afbadac842bbc58c5166962cf4f7664,
> and broke the SA_NOCLDWAIT behavior so it leaks zombies.  This fixes it.
> 
> Reported-by: Andi Kleen <andi@firstfloor.org>

Confirmed -- this patch fixes the zombie problem on my box.

Tested-by: Andi Kleen <andi@firstfloor.org>

-Andi