bce(4) sees all incoming frames as 2026 bytes in length

[Nikolay Denev]

On Apr 29, 2009, at 7:04 PM, pluknet wrote:

> 2009/4/29 Niki Denev <ndenev at gmail.com>:
>
>> bce1: <Broadcom NetXtreme II BCM5708 1000Base-T (B2)> mem
>> 0xf8000000-0xf9ffffff irq 16 at device 0.0 on pci3
>> bce1: Ethernet address: 00:22:19:xx:xx:xx
>> bce1: [ITHREAD]
>> bce1: ASIC (0x57081020); Rev (B2); Bus (PCI-X, 64-bit, 133MHz); B/C
>> (0x04040105); Flags( MFW MSI )
>>
>> bce1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0  
>> mtu 1500
>>        
>> options 
>> = 
>> 1bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,TSO4>
>>       ether 00:22:19:xx:xx:xx
>>       inet 10.18.2.1 netmask 0xffffff00 broadcast 10.18.2.255
>>       media: Ethernet autoselect (1000baseTX <full-duplex>)
>>       status: active
>>
>> And here is a tcpdump that shows the problem :
>>
>> 16:27:32.593808 00:22:19:yy:yy:yy > 00:22:19:xx:xx:xx, ethertype IPv4
>> (0x0800), length 2026: (tos 0x0, ttl 64, id 45347, offset 0, flags
>> [none], proto ICMP (1), length 84) 10.18.2.2 > 10.18.2.1: ICMP echo
>> request, id 13578, seq 36, length 64
>> 16:27:32.593817 00:22:19:xx:xx:xx > 00:22:19:yy:yy:yy, ethertype IPv4
>> (0x0800), length 98: (tos 0x0, ttl 64, id 18415, offset 0, flags
>> [none], proto ICMP (1), length 84) 10.18.2.1 > 10.18.2.2: ICMP echo
>> reply, id 13578, seq 36, length 64
>
> Ok, now I see. A link level length is 2026 for me too for some sort  
> of packets
> (in opposite to proto's len where all is ok).
>
> Mine nic is <Broadcom NetXtreme II BCM5708 1000Base-T (B2)>
> (same as yours).
>
> Looks like a regression.
> I just also tested 7.1-R and it shows expected LL-length.
>
>
> -- 
> wbr,
> pluknet


I think I got it.

It seems that the mbuf fields m_pkthdr.len and m_len are not updated  
to the real packet size pkt_len.
Well, actually they are updated, but only if we have ZERO_COPY_SOCKETS  
defined.

After I added this :

    m0->m_pkthdr.len = m0->m_len = pkt_len;

at about line 5930 in if_bce.c, the frame length reported by tcpdump  
seems correct.

P.S.: I guess this could be the cause for the lagg(4) over bce(4)  
problems too?

P.S.2: This fix will probably break the ZERO_COPY_SOCKETS case, but  
should be fairly easy to make it a "proper" fix.

Regards,
Niki Denev

-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 195 bytes
Desc: This is a digitally signed message part
Url : http://lists.freebsd.org/pipermail/freebsd-net/attachments/20090430/5217c8e7/PGP.pgp