Re: [patch 1/1] Convert the semaphore to a mutex innet/tipc/socket.c

[Andrew Morton]

On Sun, 09 Dec 2007 21:17:42 -0400
Kevin Winchester <kjwinchester@xxxxxxxxx> wrote:

> Note also that in the release method, down_interruptible() was being called
> without checking the return value.  I converted it to mutex_lock_interruptible()
> and made the interrupted case return -ERESTARTSYS, as was done for all other
> calls to down_interruptible() in the file.

That's an outright bug.

static int release(struct socket *sock)
{
	struct tipc_sock *tsock = tipc_sk(sock->sk);
	struct sock *sk = sock->sk;
	int res = TIPC_OK;
	struct sk_buff *buf;

	dbg("sock_delete: %x\n",tsock);
	if (!tsock)
		return 0;
	down_interruptible(&tsock->sem);
	if (!sock->sk) {
		up(&tsock->sem);
		return 0;
	}

	...

	up(&tsock->sem);

	...	
}

So if the calling process has signal_pending(), down_interruptible() will
return without having downed the semaphore and then we merrily proceed to
do up() on it, so a subsequent down() won't actually take the lock and
things will deteriorate from there.

So I'd propose this:

--- a/net/tipc/socket.c~a
+++ a/net/tipc/socket.c
@@ -253,7 +253,7 @@ static int release(struct socket *sock)
 	dbg("sock_delete: %x\n",tsock);
 	if (!tsock)
 		return 0;
-	down_interruptible(&tsock->sem);
+	down(&tsock->sem);
 	if (!sock->sk) {
 		up(&tsock->sem);
 		return 0;
_

as a for-2.6.24 bugfix.  And for 2.6.23.  But someone who knows what
they're doing should take a look at this...

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/