Hi Coiby,
> When two HCI_EV_CONN_COMPLETE event packets with status=0 of the same
> HCI connection are received, device_add would be called twice which
> leads to kobject_add being called twice. Thus duplicate
> (struct hci_conn *conn)->dev.kobj.entry would be inserted into
> (struct hci_conn *conn)->dev.kobj.kset->list.
>
> This issue can be fixed by checking (struct hci_conn *conn)->debugfs.
> If it's not NULL, it means the HCI connection has been completed and we
> won't duplicate the work as for processing the first
> HCI_EV_CONN_COMPLETE event.
do you have a btmon trace for this happening?
> Reported-and-tested-by: [email protected]
> Link: https://syzkaller.appspot.com/bug?extid=dd768a260f7358adbaf9
> Signed-off-by: Coiby Xu <[email protected]>
> ---
> net/bluetooth/hci_event.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
> index 4b7fc430793c..1233739ce760 100644
> --- a/net/bluetooth/hci_event.c
> +++ b/net/bluetooth/hci_event.c
> @@ -2605,6 +2605,11 @@ static void hci_conn_complete_evt(struct hci_dev *hdev, struct sk_buff *skb)
> }
>
> if (!ev->status) {
> + if (conn->debugfs) {
> + bt_dev_err(hdev, "The connection has been completed");
> + goto unlock;
> + }
> +
And instead of doing papering over a hole, I would rather detect that the HCI event is not valid since we already received one for this connection.
Regards
Marcel
On Mon, Aug 31, 2020 at 06:06:18PM +0200, Marcel Holtmann wrote:
>Hi Coiby,
Hi Marcel,
Thank you for reviewing this patch!
>
>> When two HCI_EV_CONN_COMPLETE event packets with status=0 of the same
>> HCI connection are received, device_add would be called twice which
>> leads to kobject_add being called twice. Thus duplicate
>> (struct hci_conn *conn)->dev.kobj.entry would be inserted into
>> (struct hci_conn *conn)->dev.kobj.kset->list.
>>
>> This issue can be fixed by checking (struct hci_conn *conn)->debugfs.
>> If it's not NULL, it means the HCI connection has been completed and we
>> won't duplicate the work as for processing the first
>> HCI_EV_CONN_COMPLETE event.
>
>do you have a btmon trace for this happening?
Please see the attachment "btmon_output" which is a plain text file.
I couldn't find a way to save traces in btsnoop format (the kernel would
panic immediately after running the re-producer before QEMU has a chance
to write the btsnoop file to the disk image).
I've also also attached a simplified re-producer rep9_min.c if it interests you.
>
>> Reported-and-tested-by: [email protected]
>> Link: https://syzkaller.appspot.com/bug?extid=dd768a260f7358adbaf9
>> Signed-off-by: Coiby Xu <[email protected]>
>> ---
>> net/bluetooth/hci_event.c | 5 +++++
>> 1 file changed, 5 insertions(+)
>>
>> diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
>> index 4b7fc430793c..1233739ce760 100644
>> --- a/net/bluetooth/hci_event.c
>> +++ b/net/bluetooth/hci_event.c
>> @@ -2605,6 +2605,11 @@ static void hci_conn_complete_evt(struct hci_dev *hdev, struct sk_buff *skb)
>> }
>>
>> if (!ev->status) {
>> + if (conn->debugfs) {
>> + bt_dev_err(hdev, "The connection has been completed");
>> + goto unlock;
>> + }
>> +
>
>And instead of doing papering over a hole, I would rather detect that the HCI event is not valid since we already received one for this connection.
To check conn->debugfs is what I think could be used to detect this
duplicate HCI event. Or you are suggesting this is not sufficient
and implement something like a state machine instead?
>
>Regards
>
>Marcel
>
--
Best regards,
Coiby