2007-09-22 08:02:34

by lepton

[permalink] [raw]
Subject: [RFC PATCH] 2.6.22.6 user-mode linux: before abort, we make it sure all children quit

In a stock 2.6.22.6 kernel, poweroff a user mode linux guest
(2.6.22.6 running in skas0 mode) will halt the host linux. I
think the reason is the kernel thread abort because of a bug.
Then the sys_reboot in process of user mode linux guest is
not trapped by the user mode linux kernel and is executed by host.
I think it is better to make sure all of our children process
to quit when user mode linux kernel abort.

Signed-off-by: Lepton Wu <[email protected]>

diff -X linux-2.6.22.6/Documentation/dontdiff -pru linux-2.6.22.6/arch/um/os-Linux/util.c linux-2.6.22.6-lepton/arch/um/os-Linux/util.c
--- linux-2.6.22.6/arch/um/os-Linux/util.c 2007-09-14 17:41:10.000000000 +0800
+++ linux-2.6.22.6-lepton/arch/um/os-Linux/util.c 2007-09-22 13:56:05.000000000 +0800
@@ -106,5 +106,6 @@ int setjmp_wrapper(void (*proc)(void *,
void os_dump_core(void)
{
signal(SIGSEGV, SIG_DFL);
+ kill(0, SIGTERM);
abort();
}


2007-09-25 17:53:47

by Jeff Dike

[permalink] [raw]
Subject: Re: [RFC PATCH] 2.6.22.6 user-mode linux: before abort, we make it sure all children quit

On Sat, Sep 22, 2007 at 04:01:24PM +0800, lepton wrote:
> In a stock 2.6.22.6 kernel, poweroff a user mode linux guest
> (2.6.22.6 running in skas0 mode) will halt the host linux. I
> think the reason is the kernel thread abort because of a bug.
> Then the sys_reboot in process of user mode linux guest is
> not trapped by the user mode linux kernel and is executed by host.
> I think it is better to make sure all of our children process
> to quit when user mode linux kernel abort.

Below is what I currently have for this patch. As you sent it in, the
kill(0, SIGTERM) would immediately kill the kernel process along with
everything else, before it can dump core. So, I have the kernel
ignore SIGTERM.

Then, there are still processes which survive. The one case I think I
understand is that a process is handling an infinite sequence of
SIGSEGVs and never sees the SIGTERM. So, I added a loop which waits
for all of the current child processes and kills each one as it
returns some sort of status.

Jeff

--
Work email - jdike at linux dot intel dot com

From: Lepton Wu <[email protected]>

In a stock 2.6.22.6 kernel, poweroff a user mode linux guest
(2.6.22.6 running in skas0 mode) will halt the host linux. I
think the reason is the kernel thread abort because of a bug.
Then the sys_reboot in process of user mode linux guest is
not trapped by the user mode linux kernel and is executed by host.
I think it is better to make sure all of our children process
to quit when user mode linux kernel abort.

[ jdike - the kernel process needs to ignore SIGTERM, plus the
waitpid/kill loop is needed to make sure that all of our children
are dead before the kernel exits ]

Signed-off-by: Lepton Wu <[email protected]>
Signed-off-by: Jeff Dike <[email protected]>
---
arch/um/os-Linux/skas/process.c | 2 +-
arch/um/os-Linux/util.c | 38 ++++++++++++++++++++++++++++++++++++++
2 files changed, 39 insertions(+), 1 deletion(-)

Index: linux-2.6.22/arch/um/os-Linux/util.c
===================================================================
--- linux-2.6.22.orig/arch/um/os-Linux/util.c 2007-09-25 13:33:48.000000000 -0400
+++ linux-2.6.22/arch/um/os-Linux/util.c 2007-09-25 13:45:33.000000000 -0400
@@ -105,6 +105,44 @@ int setjmp_wrapper(void (*proc)(void *,

void os_dump_core(void)
{
+ int pid;
+
signal(SIGSEGV, SIG_DFL);
+
+ /*
+ * We are about to SIGTERM this entire process group to ensure that
+ * nothing is around to run after the kernel exits. The
+ * kernel wants to abort, not die through SIGTERM, so we
+ * ignore it here.
+ */
+
+ signal(SIGTERM, SIG_IGN);
+ kill(0, SIGTERM);
+ /*
+ * Most of the other processes associated with this UML are
+ * likely sTopped, so give them a SIGCONT so they see the
+ * SIGTERM.
+ */
+ kill(0, SIGCONT);
+
+ /*
+ * Now having sent signals to everyone but us, make sure they
+ * die by ptrace. Processes can survive what's been done to
+ * them so far - the mechanism I understand is receiving a
+ * SIGSEGV and segfaulting immediately upon return. There is
+ * always a SIGSEGV pending, and (I'm guessing) signals are
+ * processed in numeric order so the SIGTERM (signal 15 vs
+ * SIGSEGV being signal 11) is never handled.
+ *
+ * Run a waitpid loop until we get some kind of error.
+ * Hopefully, it's ECHILD, but there's not a lot we can do if
+ * it's something else. Tell os_kill_ptraced_process not to
+ * wait for the child to report its death because there's
+ * nothing reasonable to do if that fails.
+ */
+
+ while ((pid = waitpid(-1, NULL, WNOHANG)) > 0)
+ os_kill_ptraced_process(pid, 0);
+
abort();
}
Index: linux-2.6.22/arch/um/os-Linux/skas/process.c
===================================================================
--- linux-2.6.22.orig/arch/um/os-Linux/skas/process.c 2007-09-25 13:34:17.000000000 -0400
+++ linux-2.6.22/arch/um/os-Linux/skas/process.c 2007-09-25 13:45:43.000000000 -0400
@@ -177,7 +177,7 @@ static int userspace_tramp(void *stack)

ptrace(PTRACE_TRACEME, 0, 0, 0);

- init_new_thread_signals();
+ signal(SIGTERM, SIG_DFL);
err = set_interval();
if (err)
panic("userspace_tramp - setting timer failed, errno = %d\n",