Re: [PATCH v2 00/15] Make the user mode driver code a better citizen

Alexei Starovoitov [off-list ref] writes:

2> On Mon, Jun 29, 2020 at 02:55:05PM -0500, Eric W. Biederman wrote:

quoted

I have tested thes changes by booting with the code compiled in and
by killing "bpfilter_umh" and running iptables -vnL to restart
the userspace driver.

I have compiled tested each change with and without CONFIG_BPFILTER
enabled.

With
CONFIG_BPFILTER=y
CONFIG_BPFILTER_UMH=m
it doesn't build:

ERROR: modpost: "kill_pid_info" [net/bpfilter/bpfilter.ko] undefined!

I've added:
+EXPORT_SYMBOL(kill_pid_info);
to continue testing...

I am rather surprised I thought Tetsuo had already compile tested
modules.

I suspect patch 13 is somehow responsible:
+	if (tgid) {
+		kill_pid_info(SIGKILL, SEND_SIG_PRIV, tgid);
+		wait_event(tgid->wait_pidfd, !pid_task(tgid, PIDTYPE_TGID));
+		bpfilter_umh_cleanup(info);
+	}

I cannot figure out why it hangs. Some sort of race ?
Since adding short delay between kill and wait makes it work.

Having had a chance to sleep kill_pid_info was a thinko, as was
!pid_task.  It should have been !pid_has_task as that takes the proper
rcu locking.

I don't know if that is going to be enough to fix the wait_event
but those are obvious bugs that need to be fixed.

diff --git a/net/bpfilter/bpfilter_kern.c b/net/bpfilter/bpfilter_kern.c
index 91474884ddb7..3e1874030daa 100644
--- a/net/bpfilter/bpfilter_kern.c
+++ b/net/bpfilter/bpfilter_kern.c

@@ -19,8 +19,8 @@ static void shutdown_umh(void)
        struct pid *tgid = info->tgid;
 
        if (tgid) {
-               kill_pid_info(SIGKILL, SEND_SIG_PRIV, tgid);
-               wait_event(tgid->wait_pidfd, !pid_task(tgid, PIDTYPE_TGID));
+               kill_pid(tgid, SIGKILL, 1);
+               wait_event(tgid->wait_pidfd, !pid_has_task(tgid, PIDTYPE_TGID));
                bpfilter_umh_cleanup(info);
        }
 }

And then did:
while true; do iptables -L;rmmod bpfilter; done
 
Unfortunately sometimes 'rmmod bpfilter' hangs in wait_event().

Hmm.  The wake up happens just of tgid->wait_pidfd happens just before
release_task is called so there is a race.  As it is possible to wake
up and then go back to sleep before pid_has_task becomes false.

So I think I need a friendly helper that does:

bool task_has_exited(struct pid *tgid)
{
	bool exited = false;

	rcu_read_lock();
        tsk = pid_task(tgid, PIDTYPE_TGID);
        exited = !!tsk;
        if (tsk) {
        	exited = !!tsk->exit_state;
out:
	rcu_unlock();
	return exited;
}

There should be a sensible way to do that.

Eric