Re: [dpdk-dev] [PATCH] net/tap: Allow all-zero checksum for UDP over IPv4

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 11/11/2020 9:31 AM, Ferruh Yigit wrote:
> On 11/11/2020 7:23 AM, Michael Pfeiffer wrote:
>> Hi,
>>
>> On Tue, 2020-11-10 at 15:59 +0000, Ferruh Yigit wrote:
>>> On 11/9/2020 2:22 PM, Michael Pfeiffer wrote:
>>>> Unlike TCP, UDP checksums are optional and may be zero to indicate "not
>>>> set" [RFC 768] (except for IPv6, where this prohibited [RFC 8200]). Add
>>>> this special case to the checksum offload emulation in net/tap.
>>>>
>>>> Signed-off-by: Michael Pfeiffer <michael.pfeiffer@tu-ilmenau.de>
>>>> ---
>>>>    drivers/net/tap/rte_eth_tap.c | 13 +++++++++++--
>>>>    1 file changed, 11 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/net/tap/rte_eth_tap.c b/drivers/net/tap/rte_eth_tap.c
>>>> index 2f8abb12c..e486b41c5 100644
>>>> --- a/drivers/net/tap/rte_eth_tap.c
>>>> +++ b/drivers/net/tap/rte_eth_tap.c
>>>> @@ -303,6 +303,7 @@ tap_verify_csum(struct rte_mbuf *mbuf)
>>>>          uint16_t cksum = 0;
>>>>          void *l3_hdr;
>>>>          void *l4_hdr;
>>>> +       struct rte_udp_hdr *udp_hdr;
>>>>          if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
>>>>                  l2_len += 4;
>>>> @@ -349,10 +350,18 @@ tap_verify_csum(struct rte_mbuf *mbuf)
>>>>                  /* Don't verify checksum for multi-segment packets. */
>>>>                  if (mbuf->nb_segs > 1)
>>>>                          return;
>>>> -               if (l3 == RTE_PTYPE_L3_IPV4)
>>>> +               if (l3 == RTE_PTYPE_L3_IPV4) {
>>>> +                       if (l4 == RTE_PTYPE_L4_UDP) {
>>>> +                               udp_hdr = (struct rte_udp_hdr *)l4_hdr;
>>>> +                               if (udp_hdr->dgram_cksum == 0) {
>>>
>>> Overall patch looks good to me, but can you please add a comment on top of
>>> above
>>> check to describe why checksum can be zero, as done in the commit log.
>>
>> Sure, I will update the patch. I am also not completely sure whether
>> PKT_RX_L4_CKSUM_NONE is the right flag for this case (rather than _UNKNOWN).
>>  From rte_core_mbuf.h:
>>
>>   * - PKT_RX_L4_CKSUM_UNKNOWN: no information about the RX L4 checksum
>>   * - PKT_RX_L4_CKSUM_NONE: the L4 checksum is not correct in the packet
>>   *   data, but the integrity of the L4 data is verified.
>>
>> The second part after the "but" is not really the case here. I don't know how
>> relevant the distinction is, as most application side code will probably only
>> do something like
>>
>> if ((mbuf->ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
>>     rte_pktmbuf_free(mbuf);
>>
>> anyway. Do you have any opinions on that?
>>
> 
> I also checked for that and wasn't sure about it :) cc'ed Olivier too for comment.
> 
> I think it is NOT 'PKT_RX_L4_CKSUM_UNKNOWN', since we know that checksum value 
> is 0x0000 which means it is not provided.
> 
> 'PKT_RX_L4_CKSUM_NONE' suits better but not sure about the expectation on 
> "integrity of the L4 data is verified" part, I assume that explanation is just 
> to differentiate between 'CKSUM_BAD'.

I suggest to continue with 'PKT_RX_L4_CKSUM_NONE'.

Can it be possible to get the new version today, so we can include this to the -rc4?

Thanks