Re: Re: Re: [PATCH V3] ax25: Fix refcount leaks caused by ax25_cb_del()

[周多明]

Hello,

On Mon, 14 Mar 2022 14:02:56 +0300, Dan Carpen wrote:

> > > But even here, my instinct is that if the refcounting is were done in
> > > the correct place we would not need any additional variables.  Is there
> > > no simpler solution?

I think there is a simpler solution instead of using any additional variables,
which is shown below:

diff --git a/net/ax25/af_ax25.c b/net/ax25/af_ax25.c
index 6bd09718077..0886109421a 100644
--- a/net/ax25/af_ax25.c
+++ b/net/ax25/af_ax25.c
@@ -98,8 +98,10 @@ static void ax25_kill_by_device(struct net_device *dev)
 			spin_unlock_bh(&ax25_list_lock);
 			lock_sock(sk);
 			s->ax25_dev = NULL;
-			dev_put_track(ax25_dev->dev, &ax25_dev->dev_tracker);
-			ax25_dev_put(ax25_dev);
+			if (sk->sk_wq) {
+				dev_put_track(ax25_dev->dev, &ax25_dev->dev_tracker);
+				ax25_dev_put(ax25_dev);
+			}
 			ax25_disconnect(s, ENETUNREACH);
 			release_sock(sk);
 			spin_lock_bh(&ax25_list_lock);
@@ -979,14 +981,20 @@ static int ax25_release(struct socket *sock)
 {
 	struct sock *sk = sock->sk;
 	ax25_cb *ax25;
+	ax25_dev *ax25_dev;
 
 	if (sk == NULL)
 		return 0;
 
 	sock_hold(sk);
-	sock_orphan(sk);
 	lock_sock(sk);
+	sock_orphan(sk);
 	ax25 = sk_to_ax25(sk);
+	ax25_dev = ax25->ax25_dev;
+	if (ax25_dev) {
+		dev_put_track(ax25_dev->dev, &ax25_dev->dev_tracker);
+		ax25_dev_put(ax25_dev);
+	}
 
 	if (sk->sk_type == SOCK_SEQPACKET) {
 		switch (ax25->state) {

we add decrements of refcounts in ax25_release(), and use lock_sock() to do synchronization. 
If refcounts decrease in ax25_release(), the decrements of refcounts in ax25_kill_by_device() 
will not be executed and vice versa. 

1. If we decrease the refcounts in ax25_release(), the decrements of refcounts in 
ax25_kill_by_device() will not execute. Because we set NULL to sk->sk_wq in sock_orphan()
and we check whether sk->sk_wq is NULL in ax25_kill_by_device(). Only positions (3) and (4)
could execute.

     (Thread 1)                      |      
ax25_bind()                          |
 ...                                 |
 ax25_addr_ax25dev()                 |
  ax25_dev_hold()   //(1)            |
  ...                                |
 dev_hold_track()   //(2)            |     (Thread 2)
 ...                                 | ax25_release()
                                     |  ...
                                     |   lock_sock(sk)
                                     |   sock_orphan(sk)
                                     |     sk->sk_wq  = NULL //set NULL
                                     |   ...
                                     |   if (ax25_dev) {
                                     |    dev_put_track() //(3)
                                     |    ax25_dev_put() //(4)
                                     |
     (thread 3)                      |
ax25_kill_by_device()                |
 ...                                 |
 lock_sock(sk)                       |
 s->ax25_dev = NULL                  |
 if (sk->sk_wq) { //check            |
  dev_put_track() //(5)              |
  ax25_dev_put() //(6)               |


2. If we decrease refcounts in ax25_kill_by_device(), the decrements of refcounts 
in ax25_release() will not execute. Because we set NULL to s->ax25_dev in ax25_kill_by_device()
and we check whether ax25_dev is NULL in ax25_release(). Only positions (3) and (4)
could execute.

     (Thread 1)                      | 
ax25_bind()                          |
 ...                                 |
 ax25_addr_ax25dev()                 |
  ax25_dev_hold()   //(1)            |
  ...                                |
 dev_hold_track()   //(2)            |      (Thread 2)
 ...                                 | ax25_kill_by_device()
                                     |  ...
                                     |   lock_sock(sk)
                                     |   s->ax25_dev = NULL //set NULL
                                     |   if (sk->sk_wq) {
                                     |    dev_put_track() //(3)
                                     |    ax25_dev_put() //(4)
                                     |    ...
                                     |
     (thread 3)                      |
ax25_release()                       |
 ...                                 |
 lock_sock(sk);                      |
 sock_orphan(sk);                    |
 ...                                 |
 if (ax25_dev) { //check             |
  dev_put_track() //(5)              |
  ax25_dev_put() //(6)               |

> > I sent "[PATCH net V2 1/2] ax25: Fix refcount leaks caused by ax25_cb_del()"
> > on On Fri, Mar 11, 2022. Could this patch solve your question?
> 
> I had a bunch of questions...  You just ignored them, and sent a patch
> called v2 instead of v4 so I was puzzled and confused. I guess the
> answer is no, could you please answer the questions?

I hope my answer could solve your questions. If you still have any questions
welcome to send email to me. I will send "[PATCH net V4 1/2] ax25: Fix
refcount leaks caused by ax25_cb_del()" as soon as possible. 

What`s more, I found NPD bugs in ax25 timers, I will send
"[PATCH net V4 2/2] ax25: Fix NULL pointer dereferences in ax25 timers" together.

Best wishes,
Duoming Zhou