Date: Mon, 20 Apr 2009 11:40:08 -0400 (EDT) From: Alan Ferrency <alan@pair.com> To: Kostik Belousov <kostikbel@gmail.com> Cc: freebsd-bugs@freebsd.org, gavin@freebsd.org, freebsd-standards@freebsd.org, Jilles Tjoelker <jilles@stack.nl> Subject: Re: bin/108390: [libc] [patch] wait4() erroneously waits for all children when SIGCHLD is SIG_IGN [regression] Message-ID: <20090420112723.S1321@smx2.pair.com> In-Reply-To: <20090419195409.GY3014@deviant.kiev.zoral.com.ua> References: <200904161242.n3GCgG4u063936@freefall.freebsd.org> <20090419195409.GY3014@deviant.kiev.zoral.com.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
> Further check on Solaris revealed that for the case where SIGCHLD is > ignored, wait4() returned -1/ECHLD, conforming to the FreeBSD behaviour > of reparenting to init. If I'm interpreting Jilles' notes about this proposed patch, it seems like this would solve our primary problem. That is: we already receive a -1/ECHLD when we call wait4pid() as expected, but we don't receive it in a timely manner if there are other child processes still running. If this patch causes wait4pid() to return immediately when a nonexistent (exited) child PID is specified, that would meet our expectations and preferences. Regarding SA_NOCLDWAIT: it was previously my understanding that SA_NOCLDWAIT and igoring SIGCHLD affect two different aspects of child process management. On a previous occasion when I looked into the behavior of SA_NOCLDWAIT, it wasn't even implemented on Linux yet; but obviously, ignoring SIGCHLD was supported. Thanks for the patch, Alan Ferrency pair Networks, Inc. > Unless further comments are given, I will commit this in several days: > > diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c > index 4255734..c35c6f2 100644 > --- a/sys/kern/kern_exit.c > +++ b/sys/kern/kern_exit.c > @@ -504,13 +504,13 @@ exit1(struct thread *td, int rv) > proc_reparent(p, initproc); > p->p_sigparent = SIGCHLD; > PROC_LOCK(p->p_pptr); > + > /* > - * If this was the last child of our parent, notify > - * parent, so in case he was wait(2)ing, he will > + * Notify parent, so in case he was wait(2)ing or > + * executiing waitpid(2) with our pid, he will > * continue. > */ > - if (LIST_EMPTY(&pp->p_children)) > - wakeup(pp); > + wakeup(pp); > } else > mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20090420112723.S1321>