Re: [PATCH] remove useless MOD_{INC,DEC}_USE_COUNT from sunrpc

[Christoph Hellwig]

On Sun, May 04, 2003 at 12:06:06PM -0700, David S. Miller wrote:
> Well, what if this hangs?  Unless the user specifies
> "--wait" to 2.5.x's rmmod, the user absolutely does not
> expect this behavior.
> 
> Rather, so that the "--wait" is respected, something one level
> up ought to be doing try_module_get().

Oh well, what about something like the following?

--- 1.25/net/sunrpc/sched.c	Thu May  1 12:52:23 2003
+++ edited/net/sunrpc/sched.c	Sun May  4 21:25:21 2003
@@ -952,6 +952,14 @@
 	wait_queue_head_t *assassin = (wait_queue_head_t*) ptr;
 	int		rounds = 0;
 
+	/*
+	 * We are locked into memory by beeing used by another module,
+	 * but the refcount might be 0 neverless so we can't use
+	 * __module_get().
+	 */
+	if (!try_module_get(THIS_MODULE))
+		return -EBUSY;
+
 	lock_kernel();
 	/*
 	 * Let our maker know we're running ...
@@ -994,6 +1002,7 @@
 
 	dprintk("RPC: rpciod exiting\n");
 	unlock_kernel();
+	module_put(THIS_MODULE);
 	return 0;
 }
 
@@ -1042,6 +1051,11 @@
 	if (error < 0) {
 		printk(KERN_WARNING "rpciod_up: create thread failed, error=%d\n", error);
 		rpciod_users--;
+		goto out;
+	}
+	if (!rpciod_pid) {
+		rpciod_users--;
+		error = -EBUSY;
 		goto out;
 	}
 	down(&rpciod_running);