Re: [PATCH v2] frame reassembly implementation for data stream

[Suraj <suraj@Atheros.com>]

Hi Marcel,

On 6/2/2010 8:32 PM, Marcel Holtmann wrote:
> Hi Suraj,
>
>> Implemented hci_recv_stream_fragment to reassemble HCI packets received from a data stream.
>>
>> Signed-off-by: suraj<suraj@Atheros.com>
>
> please fix your signed-off-by line. This is not proper.
>
>> ---
>>   include/net/bluetooth/hci_core.h |    1 +
>>   net/bluetooth/hci_core.c         |   98 ++++++++++++++++++++++++++++++++++++++
>>   2 files changed, 99 insertions(+), 0 deletions(-)
>>
>> diff --git a/include/net/bluetooth/hci_core.h b/include/net/bluetooth/hci_core.h
>> index e42f6ed..6f33f11 100644
>> --- a/include/net/bluetooth/hci_core.h
>> +++ b/include/net/bluetooth/hci_core.h
>> @@ -428,6 +428,7 @@ void hci_event_packet(struct hci_dev *hdev, struct sk_buff *skb);
>>
>>   int hci_recv_frame(struct sk_buff *skb);
>>   int hci_recv_fragment(struct hci_dev *hdev, int type, void *data, int count);
>> +int hci_recv_stream_fragment(struct hci_dev *hdev, void *data, int count);
>>
>>   int hci_register_sysfs(struct hci_dev *hdev);
>>   void hci_unregister_sysfs(struct hci_dev *hdev);
>> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
>> index 5e83f8e..ac9ccf7 100644
>> --- a/net/bluetooth/hci_core.c
>> +++ b/net/bluetooth/hci_core.c
>> @@ -1033,6 +1033,104 @@ EXPORT_SYMBOL(hci_recv_frame);
>>   /* Receive packet type fragment */
>>   #define __reassembly(hdev, type)  ((hdev)->reassembly[(type) - 2])
>>
>> +#define __get_max_rx_size(type)					\
>> +		(((type) == HCI_ACLDATA_PKT) ?			\
>> +		HCI_MAX_FRAME_SIZE :				\
>> +		((type) == HCI_EVENT_PKT) ? HCI_MAX_EVENT_SIZE :\
>> +		HCI_MAX_SCO_SIZE)
>> +
>> +#define __get_header_len(type)					\
>> +		(((type) == HCI_ACLDATA_PKT) ?			\
>> +		HCI_ACL_HDR_SIZE :				\
>> +		((type) == HCI_EVENT_PKT) ? HCI_EVENT_HDR_SIZE :\
>> +		HCI_SCO_HDR_SIZE)
>
> This is total hackish code. Who do you think is able to read this?
Will change it, thought this could make the actual function bit easier 
to read.
>
>> +int hci_recv_stream_fragment(struct hci_dev *hdev, void *data, int count)
>> +{
>> +	int type;
>> +
>> +	while (count) {
>> +		struct sk_buff *skb = __reassembly(hdev, HCI_ACLDATA_PKT);
>> +
>> +		struct { int expect; int pkt_type; } *scb;
>> +		int len = 0;
>> +
>> +		if (!skb) {
>> +			struct { char type; } *pkt;
>> +
>> +			/* Start of the frame */
>> +			pkt = data;
>> +			type = pkt->type;
>> +
>> +			if (type<  HCI_ACLDATA_PKT || type>  HCI_EVENT_PKT)
>> +				return -EILSEQ;
>> +
>> +			len = __get_max_rx_size(type);
>> +
>> +			skb = bt_skb_alloc(len, GFP_ATOMIC);
>> +			if (!skb)
>> +				return -ENOMEM;
>> +
>> +			scb = (void *) skb->cb;
>> +			scb->expect = __get_header_len(type);
>> +			scb->pkt_type = type;
>> +
>> +			skb->dev = (void *) hdev;
>> +			__reassembly(hdev, HCI_ACLDATA_PKT) = skb;
>> +
>> +			data++;
>> +			count--;
>> +
>> +			continue;
>> +		} else {
>> +			scb = (void *) skb->cb;
>> +			len = min(scb->expect, count);
>> +			type = scb->pkt_type;
>> +
>> +			memcpy(skb_put(skb, len), data, len);
>> +
>> +			count -= len;
>> +			data += len;
>> +			scb->expect -= len;
>> +		}
>> +
>> +		switch (type) {
>> +		case HCI_EVENT_PKT:
>> +			if (skb->len == HCI_EVENT_HDR_SIZE) {
>> +				struct hci_event_hdr *h = hci_event_hdr(skb);
This is a major difference. hci_recv_fragment makes a critical 
assumption that we will get the full packet header in one shot, If not 
the whole reassembly fails.
For a stream, we can not assume that. You count receive data 1 byte at 
at time. compare the detail of both function.
>> +				scb->expect = h->plen;
>> +			}
>> +			break;
>> +
>> +		case HCI_ACLDATA_PKT:
>> +			if (skb->len  == HCI_ACL_HDR_SIZE) {
>> +				struct hci_acl_hdr *h = hci_acl_hdr(skb);
>> +				scb->expect = __le16_to_cpu(h->dlen);
>> +			}
>> +			break;
>> +
>> +		case HCI_SCODATA_PKT:
>> +			if (skb->len == HCI_SCO_HDR_SIZE) {
>> +				struct hci_sco_hdr *h = hci_sco_hdr(skb);
>> +				scb->expect = h->dlen;
>> +			}
>> +			break;
>> +		}
>> +
>> +		if (scb->expect == 0) {
>> +			/* Complete frame */
>> +
>> +			__reassembly(hdev, HCI_ACLDATA_PKT) = NULL;
>> +
>> +			bt_cb(skb)->pkt_type = type;
>> +			hci_recv_frame(skb);
>> +		}
>> +
>> +	}
>> +	return 0;
>> +}
>
> I don't like this implementation at all. The biggest problem is that you
> are misusing __reassembly(hdev, HCI_ACLDATA_PKT) for getting your SKB. I
> don't wanna intermix this.
The reason why I thought about reusing this is because it might save 
another entry in the structure.

  I am also missing checks for the packet
> length matching or when packets are too big or the header size is not
> matching up.

We are dealing with a stream, not packets.
A packet can never be too big. Because, we reassemble only that number 
of bytes mentioned in packet header. The rest are part of the next 
frame.

The only option is to read the header, find length and reassemble until 
we receive the number of bytes mentioned in the header.
The rest has to be assumed as the beginning of next frame and start 
another cycle.

I guess even hci_recv_fragment works the same way.
>
> So in theory both functions do exactly the same.
  Only minor exception is
> that one knows the packet type up-front, the other has to read it from
> the stream as a 1-byte header. I don't wanna maintain two functions that
> do exactly the same.
Actually both only looks similar, but work differently. Just check the 
details.

There are certain assumptions taken by hci_recv_fragment() that need to 
be avoided.
that is we receive the packet header in one call. This can not be 
assumed for a stream.


>
> Creating an internal helper function that can maintain the current state
> of the reassembly sounds a lot better. Then re-use that function and
> ensure that the reassembly logic is inside the helper.
>
> Regards
>
> Marcel
>
>
I will work to get a better implementation. Please go through both the 
implementation and let me know your comments.


Regards
Suraj