Re: [Qemu-devel] [PATCH v4 05/11] netfilter: hook packets before net queue send

[Jason Wang <jasowang@redhat.com>]

On 08/06/2015 03:28 PM, Yang Hongyang wrote:
>
>
> On 08/06/2015 03:13 PM, Jason Wang wrote:
>>
>>
>> On 08/04/2015 04:30 PM, Yang Hongyang wrote:
>>> Capture packets that will be sent.
>>>
>>> Signed-off-by: Yang Hongyang <yanghy@cn.fujitsu.com>
>>> ---
>>>   net/net.c | 66
>>> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-
>>>   1 file changed, 65 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/net/net.c b/net/net.c
>>> index 03b2296..efccd56 100644
>>> --- a/net/net.c
>>> +++ b/net/net.c
>>> @@ -569,6 +569,42 @@ int qemu_can_send_packet(NetClientState *sender)
>>>       return 1;
>>>   }
>>>
>>> +static ssize_t filter_receive_iov(NetClientState *nc, int chain,
>>> +                                  NetClientState *sender,
>>> +                                  unsigned flags,
>>> +                                  const struct iovec *iov,
>>> +                                  int iovcnt)
>>> +{
>>> +    ssize_t ret = 0;
>>> +    NetFilterState *nf = NULL;
>>> +    ssize_t size = iov_size(iov, iovcnt);
>>> +
>>> +    QTAILQ_FOREACH(nf, &nc->filters, next) {
>>> +        if (nf->chain == chain || nf->chain == NET_FILTER_ALL) {
>>> +            ret = nf->info->receive_iov(nf, sender, flags, iov,
>>> iovcnt);
>>> +            if (ret == size) {
>>> +                return ret;
>>> +            }
>>> +        }
>>> +    }
>>> +
>>> +    return ret;
>>> +}
>>> +
>>> +static ssize_t filter_receive(NetClientState *nc, int chain,
>>> +                              NetClientState *sender,
>>> +                              unsigned flags,
>>> +                              const uint8_t *data,
>>> +                              size_t size)
>>> +{
>>> +    struct iovec iov = {
>>> +        .iov_base = (void *)data,
>>> +        .iov_len = size
>>> +    };
>>> +
>>> +    return filter_receive_iov(nc, chain, sender, flags, &iov, 1);
>>> +}
>>> +
>>>   ssize_t qemu_deliver_packet(NetClientState *sender,
>>>                               unsigned flags,
>>>                               const uint8_t *data,
>>> @@ -640,6 +676,7 @@ static ssize_t
>>> qemu_send_packet_async_with_flags(NetClientState *sender,
>>>                                                    NetPacketSent
>>> *sent_cb)
>>>   {
>>>       NetQueue *queue;
>>> +    int ret;
>>>
>>>   #ifdef DEBUG_NET
>>>       printf("qemu_send_packet_async:\n");
>>> @@ -650,6 +687,18 @@ static ssize_t
>>> qemu_send_packet_async_with_flags(NetClientState *sender,
>>>           return size;
>>>       }
>>>
>>> +    /* Let filters handle the packet first */
>>> +    ret = filter_receive(sender, NET_FILTER_OUT, sender, flags,
>>> buf, size);
>>> +    if (ret == size) {
>>> +        return size;
>>> +    }
>>> +
>>> +    ret = filter_receive(sender->peer, NET_FILTER_IN,
>>> +                         sender, flags, buf, size);
>>> +    if (ret == size) {
>>> +        return size;
>>> +    }
>>
>> I think we can just trust the return value of filter_receive(), no need
>> to check it against iov_size.
>
> You mean:
>   if (ret)
>       return ret;
> ? We need to check ret however.

Yes, but the problem is you check it against size. What if
filter_receive() returns a value which is greater than 0 but smaller
than size. If this could not happen, probably no need to check.  Not a
must but better to change this.

>
>>
>> Also sent_cb was lost track if the packet was held by filter which seems
>> wrong.
>
> So we need to extend the params of filter's receive_iov API, add a
> sent_cb
> param. 

Yes.

> BTW, it seems that sent_cb will only been called when packet is
> queueed and then sent, if the packet being sent directly sent_cb won't
> be called? am I right? or I missed somthing from the code...

Yes, but if for some reason it couldn't. Sent cb is usually used to let
main loop to re-poll network backend.

>
>>
>> [...]
>> .
>>
>