Re: [RFC PATCH v1 1/2] vsock: return errors other than -ENOMEM to socket

[Arseniy Krasnov]

On 28.03.2023 14:19, Stefano Garzarella wrote:
> On Tue, Mar 28, 2023 at 01:42:19PM +0300, Arseniy Krasnov wrote:
>>
>>
>> On 28.03.2023 12:42, Stefano Garzarella wrote:
>>> I pressed send too early...
>>>
>>> CCing Bryan, Vishnu, and pv-drivers@xxxxxxxxxx
>>>
>>> On Tue, Mar 28, 2023 at 11:39 AM Stefano Garzarella <sgarzare@xxxxxxxxxx> wrote:
>>>>
>>>> On Sun, Mar 26, 2023 at 01:13:11AM +0300, Arseniy Krasnov wrote:
>>>>> This removes behaviour, where error code returned from any transport
>>>>> was always switched to ENOMEM. This works in the same way as:
>>>>> commit
>>>>> c43170b7e157 ("vsock: return errors other than -ENOMEM to socket"),
>>>>> but for receive calls.
>>>>>
>>>>> Signed-off-by: Arseniy Krasnov <AVKrasnov@xxxxxxxxxxxxxx>
>>>>> ---
>>>>> net/vmw_vsock/af_vsock.c | 4 ++--
>>>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>>>>> index 19aea7cba26e..9262e0b77d47 100644
>>>>> --- a/net/vmw_vsock/af_vsock.c
>>>>> +++ b/net/vmw_vsock/af_vsock.c
>>>>> @@ -2007,7 +2007,7 @@ static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
>>>>>
>>>>>               read = transport->stream_dequeue(vsk, msg, len - copied, flags);
>>>>
>>>> In vmci_transport_stream_dequeue() vmci_qpair_peekv() and
>>>> vmci_qpair_dequev() return VMCI_ERROR_* in case of errors.
>>>>
>>>> Maybe we should return -ENOMEM in vmci_transport_stream_dequeue() if
>>>> those functions fail to keep the same behavior.
>>
>> Yes, seems i missed it, because several months ago we had similar question for send
>> logic:
>> https://www.spinics.net/lists/kernel/msg4611091.html
>> And it was ok to not handle VMCI send path in this way. So i think current implementation
>> for tx is a little bit buggy, because VMCI specific error from 'vmci_qpair_enquev()' is
>> returned to af_vsock.c. I think error conversion must be added to VMCI transport for tx
>> also.
> 
> Good point!
> 
> These are negative values, so there are no big problems, but I don't
> know what the user expects in this case.
> 
> @Vishnu Do we want to return an errno to the user or a VMCI_ERROR_*?

Small remark, as i can see, VMCI_ERROR_ is not exported to user in include/uapi,
so IIUC user won't be able to interpret such values correctly.

Thanks, Arseniy

> 
> In both cases I think we should do the same for both enqueue and
> dequeue.
> 
>>
>> Good thing is that Hyper-V uses general error codes.
> 
> Yeah!
> 
> Thanks,
> Stefano
> 
>>
>> Thanks, Arseniy
>>>>
>>>> CCing Bryan, Vishnu, and pv-drivers@xxxxxxxxxx
>>>>
>>>> The other transports seem okay to me.
>>>>
>>>> Thanks,
>>>> Stefano
>>>>
>>>>>               if (read < 0) {
>>>>> -                      err = -ENOMEM;
>>>>> +                      err = read;
>>>>>                       break;
>>>>>               }
>>>>>
>>>>> @@ -2058,7 +2058,7 @@ static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
>>>>>       msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
>>>>>
>>>>>       if (msg_len < 0) {
>>>>> -              err = -ENOMEM;
>>>>> +              err = msg_len;
>>>>>               goto out;
>>>>>       }
>>>>>
>>>>> -- 
>>>>> 2.25.1
>>>>>
>>>
>>
>