Re: [PATCH v4] net: rose: fix null-ptr-deref caused by rose_kill_by_neigh

[duoming]

Hello,

On Sat, 2 Jul 2022 12:01:08 -0700 Jakub Kicinski wrote:

> On Sat, 2 Jul 2022 15:23:57 +0800 (GMT+08:00) duoming@xxxxxxxxxx wrote:
> > > On Wed, 29 Jun 2022 18:49:41 +0800 Duoming Zhou wrote:  
> > > > When the link layer connection is broken, the rose->neighbour is
> > > > set to null. But rose->neighbour could be used by rose_connection()
> > > > and rose_release() later, because there is no synchronization among
> > > > them. As a result, the null-ptr-deref bugs will happen.
> > > > 
> > > > One of the null-ptr-deref bugs is shown below:
> > > > 
> > > >     (thread 1)                  |        (thread 2)
> > > >                                 |  rose_connect
> > > > rose_kill_by_neigh              |    lock_sock(sk)
> > > >   spin_lock_bh(&rose_list_lock) |    if (!rose->neighbour)
> > > >   rose->neighbour = NULL;//(1)  |
> > > >                                 |    rose->neighbour->use++;//(2)  
> > >   
> > > >  		if (rose->neighbour == neigh) {  
> > > 
> > > Why is it okay to perform this comparison without the socket lock,
> > > if we need a socket lock to clear it? Looks like rose_kill_by_neigh()
> > > is not guaranteed to clear all the uses of a neighbor.  
> > 
> > I am sorry, the comparision should also be protected with socket lock.
> > The rose_kill_by_neigh() only clear the neighbor that is passed as
> > parameter of rose_kill_by_neigh(). 
> 
> Don't think that's possible, you'd have to drop the neigh lock every
> time.

The neighbour is cleared in two situations.

(1) When the rose device is down, the rose_link_device_down() traverses
the rose_neigh_list and uses the rose_kill_by_neigh() to clear the
neighbors of the device.

void rose_link_device_down(struct net_device *dev)
{
	struct rose_neigh *rose_neigh;

	for (rose_neigh = rose_neigh_list; rose_neigh != NULL; rose_neigh = rose_neigh->next) {
		if (rose_neigh->dev == dev) {
			rose_del_route_by_neigh(rose_neigh);
			rose_kill_by_neigh(rose_neigh);
		}
	}
}

https://elixir.bootlin.com/linux/v5.19-rc4/source/net/rose/rose_route.c#L839

(2) When the level 2 link has timed out, the rose_link_failed() calls rose_kill_by_neigh()
to clear the rose_neigh.

https://elixir.bootlin.com/linux/v5.19-rc4/source/net/rose/rose_route.c#L813

> > > > +			sock_hold(s);
> > > > +			spin_unlock_bh(&rose_list_lock);
> > > > +			lock_sock(s);
> > > >  			rose_disconnect(s, ENETUNREACH, ROSE_OUT_OF_ORDER, 0);
> > > >  			rose->neighbour->use--;  
> > > 
> > > What protects the use counter?  
> > 
> > The use counter is protected by socket lock.
> 
> Which one, the neigh object can be shared by multiple sockets, no?

The sk_for_each() traverses the rose_list and uses the lock of the socket that is extracted
from the rose_list to protect the use counter.

diff --git a/net/rose/af_rose.c b/net/rose/af_rose.c
index bf2d986a6bc..6d5088b030a 100644
--- a/net/rose/af_rose.c
+++ b/net/rose/af_rose.c
@@ -165,14 +165,26 @@ void rose_kill_by_neigh(struct rose_neigh *neigh)
        struct sock *s;
 
        spin_lock_bh(&rose_list_lock);
+again:
        sk_for_each(s, &rose_list) {
                struct rose_sock *rose = rose_sk(s);
 
+               sock_hold(s);
+               spin_unlock_bh(&rose_list_lock);
+               lock_sock(s);
                if (rose->neighbour == neigh) {
                        rose_disconnect(s, ENETUNREACH, ROSE_OUT_OF_ORDER, 0);
                        rose->neighbour->use--;
                        rose->neighbour = NULL;
+                       release_sock(s);
+                       sock_put(s);
+                       spin_lock_bh(&rose_list_lock);
+                       goto again;
                }
+               release_sock(s);
+               sock_put(s);
+               spin_lock_bh(&rose_list_lock);
+               goto again;
        }
        spin_unlock_bh(&rose_list_lock);
 }

Best regards,
Duoming Zhou