On Tue, Aug 04, 2020 at 11:42:53AM +0200, Greg KH wrote:
On Tue, Aug 04, 2020 at 05:39:37PM +0800, Coiby Xu wrote:
When L2CAP channel is destroyed by hci_unregister_dev, it will
acquire the spin lock of the (struct l2cap_chan *)->tx_q list to
delete all the buffers. But sometimes when hci_unregister_dev is
being called, this lock may have not bee initialized. Initialize
the TX queue lock when creating struct l2cap_chan in 6LOWPAN to fix
this problem.
Reported-by: syzbot+fadfba6a911f6bf71842@xxxxxxxxxxxxxxxxxxxxxxxxx
Link: https://syzkaller.appspot.com/bug?extid=fadfba6a911f6bf71842
Signed-off-by: Coiby Xu <coiby.xu@xxxxxxxxx>
---
net/bluetooth/6lowpan.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/net/bluetooth/6lowpan.c b/net/bluetooth/6lowpan.c
index bb55d92691b0..713c618a73df 100644
--- a/net/bluetooth/6lowpan.c
+++ b/net/bluetooth/6lowpan.c
@@ -651,6 +651,7 @@ static struct l2cap_chan *chan_create(void)
l2cap_chan_set_defaults(chan);
+ skb_queue_head_init(&chan->tx_q);
chan->chan_type = L2CAP_CHAN_CONN_ORIENTED;
chan->mode = L2CAP_MODE_LE_FLOWCTL;
chan->imtu = 1280;
Nice, did syzbot verify that this resolves the issue?
thanks,
greg k-h
Yes. Thank you for reminding me. I'll also add an Tested-by: tag next time.
--
Best regards,
Coiby
