Hi Kumara,
IMO, TCP GSO will not use the flag. UDP GSO is the same as IP fragmentation,
and UDP fragments are treated as IP fragments. Except the first fragment, the
following fragments don’t have L4 header. RTE_PTYPE_L4_FRAG is set to
recognize them as IP fragments fast.
TCP GSO is different with IP fragmentation. All TCP segments have complete
l2/l3/l4 headers. TCP GSO doesn’t use MF flag and offset field in IP header but use
sequence number to reassemble.
The case you mentioned below is IP fragmentation, not UDP GSO or TCP GSO.
Thanks,
Jiayu
From: kumaraparameshwaran rathinavel <kumaraparamesh92@gmail.com>
Sent: Friday, June 17, 2022 7:34 PM
To: Hu, Jiayu <jiayu.hu@intel.com>
Cc: dev@dpdk.org; stable@dpdk.org
Subject: Re: [PATCH v4] gro: bug fix in identifying fragmented packets
> -----Original Message-----
> From: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
> Sent: Wednesday, June 8, 2022 5:57 PM
> To: Hu, Jiayu <jiayu.hu@intel.com>
> Cc: dev@dpdk.org; Kumara Parameshwaran
> <kumaraparamesh92@gmail.com>; stable@dpdk.org
> Subject: [PATCH v4] gro: bug fix in identifying fragmented packets
>
> From: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
>
> A packet with RTE_PTYPE_L4_FRAG(0x300) contains both RTE_PTYPE_L4_TCP
> (0x100) & RTE_PTYPE_L4_UDP (0x200). A fragmented packet as defined in
> rte_mbuf_ptype.h cannot be recognized as other L4 types and hence the
> GRO layer should not use IS_IPV4_TCP_PKT or IS_IPV4_UDP_PKT for
> RTE_PTYPE_L4_FRAG. Hence, if the packet type is RTE_PTYPE_L4_FRAG the ipA simpler way is to add a "((ptype & RTE_PTYPE_L4_FRAG) != RTE_PTYPE_L4_FRAG))"
in IS_IPV4_VXLAN_TCP4_PKT and IS_IPV4_TCP_PKT to avoid processing IP fragments
in TCP based GRO functions. For example:
#define IS_IPV4_TCP_PKT(ptype) (RTE_ETH_IS_IPV4_HDR(ptype) && \
((ptype & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP) && \
((ptype & RTE_PTYPE_L4_FRAG) != RTE_PTYPE_L4_FRAG) && \
(RTE_ETH_IS_TUNNEL_PKT(ptype) == 0))
This would be handling only the UDP fragmentation case. What if there is fragmentation of TCP packets? What happens if the packet is RTE_PTYPE_L4_FRAG and in the struct rte_ipv4_hdr the proto type is IPPROTO_TCP ? What happens to that case?
On Sun, Jun 12, 2022 at 10:50 AM Hu, Jiayu <jiayu.hu@intel.com> wrote:
Hi Kumara,
> -----Original Message-----
> From: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
> Sent: Wednesday, June 8, 2022 5:57 PM
> To: Hu, Jiayu <jiayu.hu@intel.com>
> Cc: dev@dpdk.org; Kumara Parameshwaran
> <kumaraparamesh92@gmail.com>; stable@dpdk.org
> Subject: [PATCH v4] gro: bug fix in identifying fragmented packets
>
> From: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
>
> A packet with RTE_PTYPE_L4_FRAG(0x300) contains both RTE_PTYPE_L4_TCP
> (0x100) & RTE_PTYPE_L4_UDP (0x200). A fragmented packet as defined in
> rte_mbuf_ptype.h cannot be recognized as other L4 types and hence the
> GRO layer should not use IS_IPV4_TCP_PKT or IS_IPV4_UDP_PKT for
> RTE_PTYPE_L4_FRAG. Hence, if the packet type is RTE_PTYPE_L4_FRAG the ip
A simpler way is to add a "((ptype & RTE_PTYPE_L4_FRAG) != RTE_PTYPE_L4_FRAG))"
in IS_IPV4_VXLAN_TCP4_PKT and IS_IPV4_TCP_PKT to avoid processing IP fragments
in TCP based GRO functions. For example:
#define IS_IPV4_TCP_PKT(ptype) (RTE_ETH_IS_IPV4_HDR(ptype) && \
((ptype & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP) && \
((ptype & RTE_PTYPE_L4_FRAG) != RTE_PTYPE_L4_FRAG) && \
(RTE_ETH_IS_TUNNEL_PKT(ptype) == 0))
Thanks,
Jiayu
> header should be parsed to recognize the appropriate IP type and invoke the
> respective gro handler.
>
> Fixes: 1ca5e6740852 ("gro: support UDP/IPv4")
> Cc: stable@dpdk.org
>
> Signed-off-by: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
> ---
> v1:
> * Introduce IS_IPV4_FRAGMENT macro to check if fragmented packet and
> if true extract the IP header to identify the protocol type and
> invoke the appropriate gro handler. This is done for both
> rte_gro_reassemble and rte_gro_reassemble_burst APIs.
> v2,v3,v4:
> * Fix extra whitespace and column limit warnings
>
> lib/gro/rte_gro.c | 43 +++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 41 insertions(+), 2 deletions(-) lib/gro/rte_gro.c | 43
> +++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 41 insertions(+), 2 deletions(-)
>
> diff --git a/lib/gro/rte_gro.c b/lib/gro/rte_gro.c index
> 6f7dd4d709..83d6e21dbb 100644
> --- a/lib/gro/rte_gro.c
> +++ b/lib/gro/rte_gro.c
> @@ -38,6 +38,9 @@ static gro_tbl_pkt_count_fn
> tbl_pkt_count_fn[RTE_GRO_TYPE_MAX_NUM] = {
> ((ptype & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) && \
> (RTE_ETH_IS_TUNNEL_PKT(ptype) == 0))
>
> +#define IS_IPV4_FRAGMENT(ptype) (RTE_ETH_IS_IPV4_HDR(ptype) && \
> + ((ptype & RTE_PTYPE_L4_FRAG) == RTE_PTYPE_L4_FRAG))
> +
> #define IS_IPV4_VXLAN_TCP4_PKT(ptype) (RTE_ETH_IS_IPV4_HDR(ptype)
> && \
> ((ptype & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) && \
> ((ptype & RTE_PTYPE_TUNNEL_VXLAN) == \ @@ -240,7
> +243,28 @@ rte_gro_reassemble_burst(struct rte_mbuf **pkts,
> * The timestamp is ignored, since all packets
> * will be flushed from the tables.
> */
> - if (IS_IPV4_VXLAN_TCP4_PKT(pkts[i]->packet_type) &&
> + if (IS_IPV4_FRAGMENT(pkts[i]->packet_type)) {
> + struct rte_ipv4_hdr ip4h_copy;
> + const struct rte_ipv4_hdr *ip4h =
> rte_pktmbuf_read(pkts[i], pkts[i]->l2_len,
> +
> sizeof(*ip4h), &ip4h_copy);
> + if (ip4h->next_proto_id == IPPROTO_UDP &&
> do_udp4_gro) {
> + ret = gro_udp4_reassemble(pkts[i],
> + &udp_tbl, 0);
> + if (ret > 0)
> + nb_after_gro--;
> + else if (ret < 0)
> + unprocess_pkts[unprocess_num++] =
> pkts[i];
> + } else if (ip4h->next_proto_id == IPPROTO_TCP &&
> do_tcp4_gro) {
> + ret = gro_tcp4_reassemble(pkts[i],
> + &tcp_tbl, 0);
> + if (ret > 0)
> + nb_after_gro--;
> + else if (ret < 0)
> + unprocess_pkts[unprocess_num++] =
> pkts[i];
> + } else {
> + unprocess_pkts[unprocess_num++] = pkts[i];
> + }
> + } else if (IS_IPV4_VXLAN_TCP4_PKT(pkts[i]->packet_type) &&
> do_vxlan_tcp_gro) {
> ret = gro_vxlan_tcp4_reassemble(pkts[i],
> &vxlan_tcp_tbl, 0);
> @@ -349,7 +373,22 @@ rte_gro_reassemble(struct rte_mbuf **pkts,
> current_time = rte_rdtsc();
>
> for (i = 0; i < nb_pkts; i++) {
> - if (IS_IPV4_VXLAN_TCP4_PKT(pkts[i]->packet_type) &&
> + if (IS_IPV4_FRAGMENT(pkts[i]->packet_type)) {
> + struct rte_ipv4_hdr ip4h_copy;
> + const struct rte_ipv4_hdr *ip4h =
> rte_pktmbuf_read(pkts[i], pkts[i]->l2_len,
> +
> sizeof(*ip4h), &ip4h_copy);
> + if (ip4h->next_proto_id == IPPROTO_UDP &&
> do_udp4_gro) {
> + if (gro_udp4_reassemble(pkts[i], udp_tbl,
> + current_time) < 0)
> + unprocess_pkts[unprocess_num++] =
> pkts[i];
> + } else if (ip4h->next_proto_id == IPPROTO_TCP &&
> do_tcp4_gro) {
> + if (gro_tcp4_reassemble(pkts[i], tcp_tbl,
> + current_time) < 0)
> + unprocess_pkts[unprocess_num++] =
> pkts[i];
> + } else {
> + unprocess_pkts[unprocess_num++] = pkts[i];
> + }
> + } else if (IS_IPV4_VXLAN_TCP4_PKT(pkts[i]->packet_type) &&
> do_vxlan_tcp_gro) {
> if (gro_vxlan_tcp4_reassemble(pkts[i], vxlan_tcp_tbl,
> current_time) < 0)
> --
> 2.25.1