[AX25] af_ax25: Possible circular locking.
Jarek Poplawski [Fri, 11 Jan 2008 05:21:20 +0000 (21:21 -0800)]
Bernard Pidoux F6BVP reported:
> When I killall kissattach I can see the following message.
>
> This happens on kernel 2.6.24-rc5 already patched with the 6 previously
> patches I sent recently.
>
>
> =======================================================
> [ INFO: possible circular locking dependency detected ]
> 2.6.23.9 #1
> -------------------------------------------------------
> kissattach/2906 is trying to acquire lock:
>  (linkfail_lock){-+..}, at: [<d8bd4603>] ax25_link_failed+0x11/0x39 [ax25]
>
> but task is already holding lock:
>  (ax25_list_lock){-+..}, at: [<d8bd7c7c>] ax25_device_event+0x38/0x84
> [ax25]
>
> which lock already depends on the new lock.
>
>
> the existing dependency chain (in reverse order) is:
...

lockdep is worried about the different order here:

#1 (rose_neigh_list_lock){-+..}:
#3 (ax25_list_lock){-+..}:

#0 (linkfail_lock){-+..}:
#1 (rose_neigh_list_lock){-+..}:

#3 (ax25_list_lock){-+..}:
#0 (linkfail_lock){-+..}:

So, ax25_list_lock could be taken before and after linkfail_lock.
I don't know if this three-thread clutch is very probable (or
possible at all), but it seems another bug reported by Bernard
("[...] system impossible to reboot with linux-2.6.24-rc5")
could have similar source - namely ax25_list_lock held by
ax25_kill_by_device() during ax25_disconnect(). It looks like the
only place which calls ax25_disconnect() this way, so I guess, it
isn't necessary.

This patch is breaking the lock for ax25_disconnect().

Reported-and-tested-by: Bernard Pidoux <f6bvp@free.fr>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ax25/af_ax25.c

index ecb14ee..b4725ff 100644 (file)
@@ -87,10 +87,22 @@ static void ax25_kill_by_device(struct net_device *dev)
                return;
 
        spin_lock_bh(&ax25_list_lock);
+again:
        ax25_for_each(s, node, &ax25_list) {
                if (s->ax25_dev == ax25_dev) {
                        s->ax25_dev = NULL;
+                       spin_unlock_bh(&ax25_list_lock);
                        ax25_disconnect(s, ENETUNREACH);
+                       spin_lock_bh(&ax25_list_lock);
+
+                       /* The entry could have been deleted from the
+                        * list meanwhile and thus the next pointer is
+                        * no longer valid.  Play it safe and restart
+                        * the scan.  Forward progress is ensured
+                        * because we set s->ax25_dev to NULL and we
+                        * are never passed a NULL 'dev' argument.
+                        */
+                       goto again;
                }
        }
        spin_unlock_bh(&ax25_list_lock);