Re: [PATCH net v2 1/5] vsock/virtio: discard packets if the transport changes

[Michal Luczaj <mhal@rbox.co>]

On 1/13/25 10:07, Stefano Garzarella wrote:
> On Mon, 13 Jan 2025 at 09:57, Stefano Garzarella <sgarzare@redhat.com> wrote:
>> On Sun, Jan 12, 2025 at 11:42:30PM +0100, Michal Luczaj wrote:
> 
> [...]
> 
>>>
>>> So, if I get this right:
>>> 1. vsock_create() (refcnt=1) calls vsock_insert_unbound() (refcnt=2)
>>> 2. transport->release() calls vsock_remove_bound() without checking if sk
>>>   was bound and moved to bound list (refcnt=1)
>>> 3. vsock_bind() assumes sk is in unbound list and before
>>>   __vsock_insert_bound(vsock_bound_sockets()) calls
>>>   __vsock_remove_bound() which does:
>>>      list_del_init(&vsk->bound_table); // nop
>>>      sock_put(&vsk->sk);               // refcnt=0
>>>
>>> The following fixes things for me. I'm just not certain that's the only
>>> place where transport destruction may lead to an unbound socket being
>>> removed from the unbound list.
>>>
>>> diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>>> index 7f7de6d88096..0fe807c8c052 100644
>>> --- a/net/vmw_vsock/virtio_transport_common.c
>>> +++ b/net/vmw_vsock/virtio_transport_common.c
>>> @@ -1303,7 +1303,8 @@ void virtio_transport_release(struct vsock_sock *vsk)
>>>
>>>       if (remove_sock) {
>>>               sock_set_flag(sk, SOCK_DONE);
>>> -              virtio_transport_remove_sock(vsk);
>>> +              if (vsock_addr_bound(&vsk->local_addr))
>>> +                      virtio_transport_remove_sock(vsk);
>>
>> I don't get this fix, virtio_transport_remove_sock() calls
>>    vsock_remove_sock()
>>      vsock_remove_bound()
>>        if (__vsock_in_bound_table(vsk))
>>            __vsock_remove_bound(vsk);
>>
>>
>> So, should already avoid this issue, no?
> 
> I got it wrong, I see now what are you trying to do, but I don't think
> we should skip virtio_transport_remove_sock() entirely, it also purge
> the rx_queue.

Isn't rx_queue empty-by-definition in case of !__vsock_in_bound_table(vsk)?

>> Can the problem be in vsock_bind() ?

Well, I wouldn't say so.

>> Is this issue pre-existing or introduced by this series?
> 
> I think this is pre-existing, can you confirm?

Yup, I agree, pre-existing.

> In that case, I'd not stop this series, and fix it in another patch/series.

Yeah, sure thing.

Thanks,
Michal