Re: [PATCH RFC net v2 2/3] vsock: Fix transport_* TOCTOU

[Michal Luczaj <mhal@rbox.co>]

On 6/27/25 10:08, Stefano Garzarella wrote:
> On Fri, Jun 20, 2025 at 09:52:44PM +0200, Michal Luczaj wrote:
>> Transport assignment may race with module unload. Protect new_transport
>>from becoming a stale pointer.
>>
>> This also takes care of an insecure call in vsock_use_local_transport();
>> add a lockdep assert.
>>
>> BUG: unable to handle page fault for address: fffffbfff8056000
>> Oops: Oops: 0000 [#1] SMP KASAN
>> RIP: 0010:vsock_assign_transport+0x366/0x600
>> Call Trace:
>> vsock_connect+0x59c/0xc40
>> __sys_connect+0xe8/0x100
>> __x64_sys_connect+0x6e/0xc0
>> do_syscall_64+0x92/0x1c0
>> entry_SYSCALL_64_after_hwframe+0x4b/0x53
>>
>> Fixes: c0cfa2d8a788 ("vsock: add multi-transports support")
>> Signed-off-by: Michal Luczaj <mhal@rbox.co>
>> ---
>> net/vmw_vsock/af_vsock.c | 28 +++++++++++++++++++++++-----
>> 1 file changed, 23 insertions(+), 5 deletions(-)
>>
>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>> index 63a920af5bfe6960306a3e5eeae0cbf30648985e..a1b1073a2c89f865fcdb58b38d8e7feffcf1544f 100644
>> --- a/net/vmw_vsock/af_vsock.c
>> +++ b/net/vmw_vsock/af_vsock.c
>> @@ -407,6 +407,8 @@ EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
>>
>> static bool vsock_use_local_transport(unsigned int remote_cid)
>> {
>> +	lockdep_assert_held(&vsock_register_mutex);
>> +
>> 	if (!transport_local)
>> 		return false;
>>
>> @@ -464,6 +466,8 @@ int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
>>
>> 	remote_flags = vsk->remote_addr.svm_flags;
>>
>> +	mutex_lock(&vsock_register_mutex);
>> +
>> 	switch (sk->sk_type) {
>> 	case SOCK_DGRAM:
>> 		new_transport = transport_dgram;
>> @@ -479,12 +483,15 @@ int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
>> 			new_transport = transport_h2g;
>> 		break;
>> 	default:
>> -		return -ESOCKTNOSUPPORT;
>> +		ret = -ESOCKTNOSUPPORT;
>> +		goto err;
>> 	}
>>
>> 	if (vsk->transport) {
>> -		if (vsk->transport == new_transport)
>> -			return 0;
>> +		if (vsk->transport == new_transport) {
>> +			ret = 0;
>> +			goto err;
>> +		}
> 
> 		/* transport->release() must be called with sock lock acquired.
> 		 * This path can only be taken during vsock_connect(), where we
> 		 * have already held the sock lock. In the other cases, this
> 		 * function is called on a new socket which is not assigned to
> 		 * any transport.
> 		 */
> 		vsk->transport->release(vsk);
> 		vsock_deassign_transport(vsk);
> 
> Thinking back to this patch, could there be a deadlock between call
> vsock_deassign_transport(), which call module_put(), now with the
> `vsock_register_mutex` held, and the call to vsock_core_unregister()
> usually made by modules in the exit function?

I think we're good. module_put() does not call the module cleanup function
(kernel/module/main.c:delete_module() syscall does that), so
vsock_core_unregister() won't happen in this path here. Have I missed
anything else?