bpfilter: fix race in pipe access

syzbot reported the following crash
[  338.293946] bpfilter: read fail -512
[  338.304515] kasan: GPF could be caused by NULL-ptr deref or user memory access
[  338.311863] general protection fault: 0000 [#1] SMP KASAN
[  338.344360] RIP: 0010:__vfs_write+0x4a6/0x960
[  338.426363] Call Trace:
[  338.456967]  __kernel_write+0x10c/0x380
[  338.460928]  __bpfilter_process_sockopt+0x1d8/0x35b
[  338.487103]  bpfilter_mbox_request+0x4d/0xb0
[  338.491492]  bpfilter_ip_get_sockopt+0x6b/0x90

This can happen when multiple cpus trying to talk to user mode process
via bpfilter_mbox_request(). One cpu grabs the mutex while another goes to
sleep on the same mutex. Then former cpu sees that umh pipe is down and
shuts down the pipes. Later cpu finally acquires the mutex and crashes
on freed pipe.
Fix the race by using info.pid as an indicator that umh and pipes are healthy
and check it after acquiring the mutex.

Fixes: d2ba09c ("net: add skeleton of bpfilter kernel module")
Reported-by: [email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Author: Alexei Starovoitov

net/bpfilter/bpfilter_kern.c

@@ -24,17 +24,19 @@ static void shutdown_umh(struct umh_info *info)
 {
 	struct task_struct *tsk;
 
+	if (!info->pid)
+		return;
 	tsk = pid_task(find_vpid(info->pid), PIDTYPE_PID);
 	if (tsk)
 		force_sig(SIGKILL, tsk);
 	fput(info->pipe_to_umh);
 	fput(info->pipe_from_umh);
+	info->pid = 0;
 }
 
 static void __stop_umh(void)
 {
-	if (IS_ENABLED(CONFIG_INET) &&
-	    bpfilter_process_sockopt) {
+	if (IS_ENABLED(CONFIG_INET)) {
 		bpfilter_process_sockopt = NULL;
 		shutdown_umh(&info);
 	}
@@ -55,14 +57,16 @@ static int __bpfilter_process_sockopt(struct sock *sk, int optname,
 	struct mbox_reply reply;
 	loff_t pos;
 	ssize_t n;
-	int ret;
+	int ret = -EFAULT;
 
 	req.is_set = is_set;
 	req.pid = current->pid;
 	req.cmd = optname;
 	req.addr = (long)optval;
 	req.len = optlen;
 	mutex_lock(&bpfilter_lock);
+	if (!info.pid)
+		goto out;
 	n = __kernel_write(info.pipe_to_umh, &req, sizeof(req), &pos);
 	if (n != sizeof(req)) {
 		pr_err("write fail %zd\n", n);