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

[Michael Pfeiffer <michael.pfeiffer@tu-ilmenau.de>]

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?

> > +                                       mbuf->ol_flags |=
> > PKT_RX_L4_CKSUM_NONE;
> > +                                       return;
> > +                               }
> > +                       }
> >                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
> > -               else if (l3 == RTE_PTYPE_L3_IPV6)
> > +               } else if (l3 == RTE_PTYPE_L3_IPV6) {
> >                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
> > +               }
> >                 mbuf->ol_flags |= cksum ?
> >                         PKT_RX_L4_CKSUM_BAD :
> >                         PKT_RX_L4_CKSUM_GOOD;
> > 
> 

Regards
Michael