[ovs-dev,v2,1/2] userspace datapath: Add GTP-U tunnel support

From: Feng Yang <feng.yang@intel.com>

GPRS Tunneling Protocol (GTP) is a group of IP-based communications
protocols used to carry general packet radio service (GPRS) within
GSM, UMTS and LTE networks.

GTP can be decomposed into separate protocols, GTP-C, GTP-U and GTP'.

GTP-U is used for carrying user data within the GPRS core network and
between the radio access network and the core network. The user data
transported can be packets in any of IPv4, IPv6, or PPP formats.

GTP-U is user plane communication protocol between Serving-Gateway
(S-GW) and PDN-Gateway (P-GW) as well as Multi-access Edge Computing
(MEC).

This patch added a new tunnel type 'gtpu' into userspace datapath and
two new match fields tun_gtpu_flags and tun_gtpu_msgtype, tun_id is
reused for GTP Tunnel Endpoint ID (TEID).

The below are commands for gtpu tunnel add:

$ sudo ovs-vsctl add-br br-int -- set bridge br-int datapath_type=netdev protocols=OpenFlow13
$ sudo ovs-vsctl add-port br-int gtpu1 -- set interface gtpu1 type=gtpu options:packet_type=legacy_l3 options:remote_ip=flow options:key=flow options:dst_port=2152

test-flows-gtpu.txt is openflow table for gtpu tunnel test:

$ cat test-flows-gtpu.txt
table=0,icmp,in_port=4 actions=load:0xc0a83249->NXM_NX_TUN_IPV4_DST[],load:0x9->NXM_NX_TUN_ID[0..31],output:2
table=0,tcp,tp_dst=80,in_port=4 actions=load:0xc0a83249->NXM_NX_TUN_IPV4_DST[],load:0x9->NXM_NX_TUN_ID[0..31],output:2
table=0,in_port=2,tun_id=0x10,tun_gtpu_flags=0x30,tun_gtpu_msgtype=255 actions=set_field:00:00:22:22:22:22->eth_src,set_field:00:00:11:11:11:11->eth_dst,output:4
table=0,actions=NORMAL

Signed-off-by: Feng Yang <feng.yang@intel.com>
Signed-off-by: Jiannan Ouyang <ouyangj@fb.com>
Signed-off-by: Yi Yang <yi.y.yang@intel.com>
---
 build-aux/extract-ofp-fields                      |   1 +
 datapath/linux/compat/include/linux/openvswitch.h |   3 +
 include/openvswitch/match.h                       |   4 +
 include/openvswitch/meta-flow.h                   |  28 +++++
 include/openvswitch/packets.h                     |   4 +-
 lib/dpif-netlink-rtnl.c                           |   5 +
 lib/dpif-netlink.c                                |   5 +
 lib/flow.c                                        |   7 ++
 lib/match.c                                       |  29 +++++
 lib/meta-flow.c                                   |  40 ++++++
 lib/meta-flow.xml                                 |  76 ++++++++++++
 lib/netdev-native-tnl.c                           | 145 +++++++++++++++++++++-
 lib/netdev-native-tnl.h                           |  12 ++
 lib/netdev-vport.c                                |  18 ++-
 lib/nx-match.c                                    |   4 +
 lib/odp-util.c                                    |  48 +++++++
 lib/packets.h                                     |  19 +++
 lib/tnl-ports.c                                   |   3 +
 ofproto/ofproto-dpif-ipfix.c                      |   5 +
 ofproto/ofproto-dpif-sflow.c                      |   6 +-
 ofproto/ofproto-dpif-xlate.c                      |   4 +-
 ofproto/tunnel.c                                  |   9 +-
 vswitchd/vswitch.xml                              |  16 +++
 23 files changed, 485 insertions(+), 6 deletions(-)

Yi Yang <yi.y.yang@intel.com> writes:

Hi Yi,

Thanks for the patch!  Just a brief review.

> From: Feng Yang <feng.yang@intel.com>
>
> GPRS Tunneling Protocol (GTP) is a group of IP-based communications
> protocols used to carry general packet radio service (GPRS) within
> GSM, UMTS and LTE networks.
>
> GTP can be decomposed into separate protocols, GTP-C, GTP-U and GTP'.
>
> GTP-U is used for carrying user data within the GPRS core network and
> between the radio access network and the core network. The user data
> transported can be packets in any of IPv4, IPv6, or PPP formats.
>
> GTP-U is user plane communication protocol between Serving-Gateway
> (S-GW) and PDN-Gateway (P-GW) as well as Multi-access Edge Computing
> (MEC).
>
> This patch added a new tunnel type 'gtpu' into userspace datapath and
> two new match fields tun_gtpu_flags and tun_gtpu_msgtype, tun_id is
> reused for GTP Tunnel Endpoint ID (TEID).
>
> The below are commands for gtpu tunnel add:
>
> $ sudo ovs-vsctl add-br br-int -- set bridge br-int datapath_type=netdev protocols=OpenFlow13
> $ sudo ovs-vsctl add-port br-int gtpu1 -- set interface gtpu1 type=gtpu options:packet_type=legacy_l3 options:remote_ip=flow options:key=flow options:dst_port=2152
>
> test-flows-gtpu.txt is openflow table for gtpu tunnel test:
>
> $ cat test-flows-gtpu.txt
> table=0,icmp,in_port=4 actions=load:0xc0a83249->NXM_NX_TUN_IPV4_DST[],load:0x9->NXM_NX_TUN_ID[0..31],output:2
> table=0,tcp,tp_dst=80,in_port=4 actions=load:0xc0a83249->NXM_NX_TUN_IPV4_DST[],load:0x9->NXM_NX_TUN_ID[0..31],output:2
> table=0,in_port=2,tun_id=0x10,tun_gtpu_flags=0x30,tun_gtpu_msgtype=255 actions=set_field:00:00:22:22:22:22->eth_src,set_field:00:00:11:11:11:11->eth_dst,output:4
> table=0,actions=NORMAL
>
> Signed-off-by: Feng Yang <feng.yang@intel.com>
> Signed-off-by: Jiannan Ouyang <ouyangj@fb.com>
> Signed-off-by: Yi Yang <yi.y.yang@intel.com>

Needs Co-author tags.

> ---
>  build-aux/extract-ofp-fields                      |   1 +
>  datapath/linux/compat/include/linux/openvswitch.h |   3 +
>  include/openvswitch/match.h                       |   4 +
>  include/openvswitch/meta-flow.h                   |  28 +++++
>  include/openvswitch/packets.h                     |   4 +-
>  lib/dpif-netlink-rtnl.c                           |   5 +
>  lib/dpif-netlink.c                                |   5 +
>  lib/flow.c                                        |   7 ++
>  lib/match.c                                       |  29 +++++
>  lib/meta-flow.c                                   |  40 ++++++
>  lib/meta-flow.xml                                 |  76 ++++++++++++
>  lib/netdev-native-tnl.c                           | 145 +++++++++++++++++++++-
>  lib/netdev-native-tnl.h                           |  12 ++
>  lib/netdev-vport.c                                |  18 ++-
>  lib/nx-match.c                                    |   4 +
>  lib/odp-util.c                                    |  48 +++++++
>  lib/packets.h                                     |  19 +++
>  lib/tnl-ports.c                                   |   3 +
>  ofproto/ofproto-dpif-ipfix.c                      |   5 +
>  ofproto/ofproto-dpif-sflow.c                      |   6 +-
>  ofproto/ofproto-dpif-xlate.c                      |   4 +-
>  ofproto/tunnel.c                                  |   9 +-
>  vswitchd/vswitch.xml                              |  16 +++
>  23 files changed, 485 insertions(+), 6 deletions(-)
>
> diff --git a/build-aux/extract-ofp-fields b/build-aux/extract-ofp-fields
> index 184b75e..944f095 100755
> --- a/build-aux/extract-ofp-fields
> +++ b/build-aux/extract-ofp-fields
> @@ -68,6 +68,7 @@ PREREQS = {"none": "MFP_NONE",
>  OXM_CLASSES = {"NXM_OF_":        (0,          0x0000),
>                 "NXM_NX_":        (0,          0x0001),
>                 "NXOXM_NSH_":     (0x005ad650, 0xffff),
> +               "NXOXM_GTPU_":    (0x005ad651, 0xffff),
>                 "OXM_OF_":        (0,          0x8000),
>                 "OXM_OF_PKT_REG": (0,          0x8001),
>                 "ONFOXM_ET_":     (0x4f4e4600, 0xffff),
> diff --git a/datapath/linux/compat/include/linux/openvswitch.h b/datapath/linux/compat/include/linux/openvswitch.h
> index 84ebcaf..ad6ea64 100644
> --- a/datapath/linux/compat/include/linux/openvswitch.h
> +++ b/datapath/linux/compat/include/linux/openvswitch.h

I don't see this in the upstream code.  This is introducing another
place for kernel and ovs to diverge.  Please get this type accepted
upstream first.

> @@ -236,6 +236,7 @@ enum ovs_vport_type {
>  	OVS_VPORT_TYPE_GRE,      /* GRE tunnel. */
>  	OVS_VPORT_TYPE_VXLAN,	 /* VXLAN tunnel. */
>  	OVS_VPORT_TYPE_GENEVE,	 /* Geneve tunnel. */
> +	OVS_VPORT_TYPE_GTPU,     /* GTPU tunnel. */
>  	OVS_VPORT_TYPE_LISP = 105,  /* LISP tunnel */
>  	OVS_VPORT_TYPE_STT = 106, /* STT tunnel */
>  	__OVS_VPORT_TYPE_MAX
> @@ -394,6 +395,8 @@ enum ovs_tunnel_key_attr {
>  	OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS,		/* Nested OVS_VXLAN_EXT_* */
>  	OVS_TUNNEL_KEY_ATTR_IPV6_SRC,		/* struct in6_addr src IPv6 address. */
>  	OVS_TUNNEL_KEY_ATTR_IPV6_DST,		/* struct in6_addr dst IPv6 address. */
> +	OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS,         /* u8 GTP-U FLAGS. */
> +	OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE,       /* u8 GTP-U MSGTYPE. */
>  	OVS_TUNNEL_KEY_ATTR_PAD,
>  	__OVS_TUNNEL_KEY_ATTR_MAX
>  };
> diff --git a/include/openvswitch/match.h b/include/openvswitch/match.h
> index 49333ae..9bcc4f4 100644
> --- a/include/openvswitch/match.h
> +++ b/include/openvswitch/match.h
> @@ -109,6 +109,10 @@ void match_set_tun_gbp_id_masked(struct match *match, ovs_be16 gbp_id, ovs_be16
>  void match_set_tun_gbp_id(struct match *match, ovs_be16 gbp_id);
>  void match_set_tun_gbp_flags_masked(struct match *match, uint8_t flags, uint8_t mask);
>  void match_set_tun_gbp_flags(struct match *match, uint8_t flags);
> +void match_set_tun_gtpu_flags(struct match *match, uint8_t gtpu_flags);
> +void match_set_tun_gtpu_flags_masked(struct match *match, uint8_t gtpu_flags,
> +                                     uint8_t mask);
> +void match_set_tun_gtpu_msgtype(struct match *match, uint8_t gtpu_msgtype);
>  void match_set_in_port(struct match *, ofp_port_t ofp_port);
>  void match_set_pkt_mark(struct match *, uint32_t pkt_mark);
>  void match_set_pkt_mark_masked(struct match *, uint32_t pkt_mark, uint32_t mask);
> diff --git a/include/openvswitch/meta-flow.h b/include/openvswitch/meta-flow.h
> index 98c9e1c..884a2f6 100644
> --- a/include/openvswitch/meta-flow.h
> +++ b/include/openvswitch/meta-flow.h
> @@ -1848,6 +1848,34 @@ enum OVS_PACKED_ENUM mf_field_id {
>       */
>      MFF_NSH_TTL,
>  
> +    /* "tun_gtpu_flags".
> +     *
> +     * The flags in a GTP-U tunnel header.
> +     *
> +     * Type: u8.
> +     * Maskable: bitwise.
> +     * Formatting: decimal.
> +     * Prerequisites: none.
> +     * Access: read/write.
> +     * NXM: none.
> +     * OXM: NXOXM_GTPU_FLAGS(1) since OF1.3 and v2.10.
> +     */
> +    MFF_TUN_GTPU_FLAGS,
> +
> +    /* "tun_gtpu_msgtype".
> +     *
> +     * The message type in a GTP-U tunnel header.
> +     *
> +     * Type: u8.
> +     * Maskable: no.
> +     * Formatting: decimal.
> +     * Prerequisites: none.
> +     * Access: read/write.
> +     * NXM: none.
> +     * OXM: NXOXM_GTPU_MSGTYPE(2) since OF1.3 and v2.10.
> +     */
> +    MFF_TUN_GTPU_MSGTYPE,
> +
>      MFF_N_IDS
>  };
>  
> diff --git a/include/openvswitch/packets.h b/include/openvswitch/packets.h
> index fef756b..1f3b499 100644
> --- a/include/openvswitch/packets.h
> +++ b/include/openvswitch/packets.h
> @@ -39,7 +39,9 @@ struct flow_tnl {
>      ovs_be16 tp_dst;
>      ovs_be16 gbp_id;
>      uint8_t  gbp_flags;
> -    uint8_t  pad1[5];        /* Pad to 64 bits. */
> +    uint8_t  gtpu_flags;
> +    uint8_t  gtpu_msgtype;
> +    uint8_t  pad1[3];        /* Pad to 64 bits. */
>      struct tun_metadata metadata;
>  };
>  
> diff --git a/lib/dpif-netlink-rtnl.c b/lib/dpif-netlink-rtnl.c
> index 40c4569..a628b72 100644
> --- a/lib/dpif-netlink-rtnl.c
> +++ b/lib/dpif-netlink-rtnl.c

Again, you'll need to get the netlink datapath changes through kernel
process.  At least try.

> @@ -99,6 +99,8 @@ vport_type_to_kind(enum ovs_vport_type type,
>          }
>      case OVS_VPORT_TYPE_GENEVE:
>          return "geneve";
> +    case OVS_VPORT_TYPE_GTPU:
> +        return "gtpu";
>      case OVS_VPORT_TYPE_NETDEV:
>      case OVS_VPORT_TYPE_INTERNAL:
>      case OVS_VPORT_TYPE_LISP:
> @@ -262,6 +264,7 @@ dpif_netlink_rtnl_verify(const struct netdev_tunnel_config *tnl_cfg,
>      case OVS_VPORT_TYPE_INTERNAL:
>      case OVS_VPORT_TYPE_LISP:
>      case OVS_VPORT_TYPE_STT:
> +    case OVS_VPORT_TYPE_GTPU:
>      case OVS_VPORT_TYPE_UNSPEC:
>      case __OVS_VPORT_TYPE_MAX:
>      default:
> @@ -327,6 +330,7 @@ dpif_netlink_rtnl_create(const struct netdev_tunnel_config *tnl_cfg,
>      case OVS_VPORT_TYPE_INTERNAL:
>      case OVS_VPORT_TYPE_LISP:
>      case OVS_VPORT_TYPE_STT:
> +    case OVS_VPORT_TYPE_GTPU:
>      case OVS_VPORT_TYPE_UNSPEC:
>      case __OVS_VPORT_TYPE_MAX:
>      default:
> @@ -438,6 +442,7 @@ dpif_netlink_rtnl_port_destroy(const char *name, const char *type)
>      case OVS_VPORT_TYPE_INTERNAL:
>      case OVS_VPORT_TYPE_LISP:
>      case OVS_VPORT_TYPE_STT:
> +    case OVS_VPORT_TYPE_GTPU:
>      case OVS_VPORT_TYPE_UNSPEC:
>      case __OVS_VPORT_TYPE_MAX:
>      default:
> diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
> index bb9e95d..8b5bd97 100644
> --- a/lib/dpif-netlink.c
> +++ b/lib/dpif-netlink.c
> @@ -787,6 +787,9 @@ get_vport_type(const struct dpif_netlink_vport *vport)
>      case OVS_VPORT_TYPE_STT:
>          return "stt";
>  
> +    case OVS_VPORT_TYPE_GTPU:
> +        return "gtpu";
> +
>      case OVS_VPORT_TYPE_UNSPEC:
>      case __OVS_VPORT_TYPE_MAX:
>          break;
> @@ -814,6 +817,8 @@ netdev_to_ovs_vport_type(const char *type)
>          return OVS_VPORT_TYPE_VXLAN;
>      } else if (!strcmp(type, "lisp")) {
>          return OVS_VPORT_TYPE_LISP;
> +    } else if (strstr(type, "gtpu")) {

Why isn't this using "!strcmp"?

> +        return OVS_VPORT_TYPE_GTPU;
>      } else {
>          return OVS_VPORT_TYPE_UNSPEC;
>      }
> diff --git a/lib/flow.c b/lib/flow.c
> index 09b66b8..d5af1e8 100644
> --- a/lib/flow.c
> +++ b/lib/flow.c
> @@ -695,6 +695,13 @@ miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
>          miniflow_push_be32(mf, packet_type, packet_type);
>      }
>  
> +    /* It won't be parsed if the packet contains application layer data only.
> +     * These include, but not limited to GTP-U messages, GTP-C packets.
> +     */

Why treat GTP-U and C separate here?

> +    if (packet_type == htonl(PT_GTPU_MSG)) {
> +        goto out;
> +    }
> +
>      /* Initialize packet's layer pointer and offsets. */
>      frame = data;
>      dp_packet_reset_offsets(packet);
> diff --git a/lib/match.c b/lib/match.c
> index 2e9a803..e5cc9c6 100644
> --- a/lib/match.c
> +++ b/lib/match.c
> @@ -320,6 +320,27 @@ match_set_tun_gbp_flags(struct match *match, uint8_t flags)
>  }
>  
>  void
> +match_set_tun_gtpu_flags(struct match *match, uint8_t gtpu_flags)
> +{
> +    match_set_tun_gtpu_flags_masked(match, gtpu_flags, UINT8_MAX);
> +}
> +
> +void
> +match_set_tun_gtpu_flags_masked(struct match *match, uint8_t gtpu_flags,
> +                                uint8_t mask)
> +{
> +    match->wc.masks.tunnel.gtpu_flags = mask;
> +    match->flow.tunnel.gtpu_flags = gtpu_flags & mask;
> +}
> +
> +void
> +match_set_tun_gtpu_msgtype(struct match *match, uint8_t gtpu_msgtype)
> +{
> +    match->flow.tunnel.gtpu_msgtype = gtpu_msgtype;
> +    match->wc.masks.tunnel.gtpu_msgtype = UINT8_MAX;
> +}
> +
> +void
>  match_set_in_port(struct match *match, ofp_port_t ofp_port)
>  {
>      match->wc.masks.in_port.ofp_port = u16_to_ofp(UINT16_MAX);
> @@ -1239,6 +1260,14 @@ format_flow_tunnel(struct ds *s, const struct match *match)
>                              FLOW_TNL_F_MASK);
>          ds_put_char(s, ',');
>      }
> +
> +    if (wc->masks.tunnel.gtpu_flags) {
> +        ds_put_format(s, "tun_gtpu_flags=%"PRIx8",", tnl->gtpu_flags);
> +    }
> +    if (wc->masks.tunnel.gtpu_msgtype) {
> +        ds_put_format(s, "tun_gtpu_msgtype=%"PRIu8",", tnl->gtpu_msgtype);
> +    }
> +
>      tun_metadata_match_format(s, match);
>  }
>  
> diff --git a/lib/meta-flow.c b/lib/meta-flow.c
> index aa2ec01..bde59df 100644
> --- a/lib/meta-flow.c
> +++ b/lib/meta-flow.c
> @@ -231,6 +231,10 @@ mf_is_all_wild(const struct mf_field *mf, const struct flow_wildcards *wc)
>          return !wc->masks.tunnel.gbp_id;
>      case MFF_TUN_GBP_FLAGS:
>          return !wc->masks.tunnel.gbp_flags;
> +    case MFF_TUN_GTPU_FLAGS:
> +        return !wc->masks.tunnel.gtpu_flags;
> +    case MFF_TUN_GTPU_MSGTYPE:
> +        return !wc->masks.tunnel.gtpu_msgtype;
>      CASE_MFF_TUN_METADATA:
>          return !ULLONG_GET(wc->masks.tunnel.metadata.present.map,
>                             mf->id - MFF_TUN_METADATA0);
> @@ -513,6 +517,7 @@ mf_is_value_valid(const struct mf_field *mf, const union mf_value *value)
>      case MFF_TUN_TTL:
>      case MFF_TUN_GBP_ID:
>      case MFF_TUN_GBP_FLAGS:
> +    case MFF_TUN_GTPU_MSGTYPE:
>      CASE_MFF_TUN_METADATA:
>      case MFF_METADATA:
>      case MFF_IN_PORT:
> @@ -625,6 +630,9 @@ mf_is_value_valid(const struct mf_field *mf, const union mf_value *value)
>      case MFF_NSH_C4:
>          return true;
>  
> +    case MFF_TUN_GTPU_FLAGS:
> +        return (value->u8 == GTPU_FLAGS_DEFAULT);
> +
>      case MFF_N_IDS:
>      default:
>          OVS_NOT_REACHED();
> @@ -674,6 +682,12 @@ mf_get_value(const struct mf_field *mf, const struct flow *flow,
>      case MFF_TUN_GBP_FLAGS:
>          value->u8 = flow->tunnel.gbp_flags;
>          break;
> +    case MFF_TUN_GTPU_FLAGS:
> +        value->u8 = flow->tunnel.gtpu_flags;
> +        break;
> +    case MFF_TUN_GTPU_MSGTYPE:
> +        value->u8 = flow->tunnel.gtpu_msgtype;
> +        break;
>      case MFF_TUN_TTL:
>          value->u8 = flow->tunnel.ip_ttl;
>          break;
> @@ -988,6 +1002,12 @@ mf_set_value(const struct mf_field *mf,
>      case MFF_TUN_GBP_FLAGS:
>           match_set_tun_gbp_flags(match, value->u8);
>           break;
> +    case MFF_TUN_GTPU_FLAGS:
> +        match_set_tun_gtpu_flags(match, value->u8);
> +        break;
> +    case MFF_TUN_GTPU_MSGTYPE:
> +        match_set_tun_gtpu_msgtype(match, value->u8);
> +        break;
>      case MFF_TUN_TOS:
>          match_set_tun_tos(match, value->u8);
>          break;
> @@ -1385,6 +1405,12 @@ mf_set_flow_value(const struct mf_field *mf,
>      case MFF_TUN_GBP_FLAGS:
>          flow->tunnel.gbp_flags = value->u8;
>          break;
> +    case MFF_TUN_GTPU_FLAGS:
> +        flow->tunnel.gtpu_flags = value->u8;
> +        break;
> +    case MFF_TUN_GTPU_MSGTYPE:
> +        flow->tunnel.gtpu_msgtype = value->u8;
> +        break;
>      case MFF_TUN_TOS:
>          flow->tunnel.ip_tos = value->u8;
>          break;
> @@ -1700,6 +1726,8 @@ mf_is_pipeline_field(const struct mf_field *mf)
>      case MFF_TUN_FLAGS:
>      case MFF_TUN_GBP_ID:
>      case MFF_TUN_GBP_FLAGS:
> +    case MFF_TUN_GTPU_FLAGS:
> +    case MFF_TUN_GTPU_MSGTYPE:
>      CASE_MFF_TUN_METADATA:
>      case MFF_METADATA:
>      case MFF_IN_PORT:
> @@ -1870,6 +1898,14 @@ mf_set_wild(const struct mf_field *mf, struct match *match, char **err_str)
>      case MFF_TUN_GBP_FLAGS:
>          match_set_tun_gbp_flags_masked(match, 0, 0);
>          break;
> +    case MFF_TUN_GTPU_FLAGS:
> +        match->wc.masks.tunnel.gtpu_flags = 0;
> +        match->flow.tunnel.gtpu_flags = 0;
> +        break;
> +    case MFF_TUN_GTPU_MSGTYPE:
> +        match->wc.masks.tunnel.gtpu_msgtype = 0;
> +        match->flow.tunnel.gtpu_msgtype = 0;
> +        break;
>      case MFF_TUN_TOS:
>          match_set_tun_tos_masked(match, 0, 0);
>          break;
> @@ -2221,6 +2257,7 @@ mf_set(const struct mf_field *mf,
>      case MFF_ICMPV4_CODE:
>      case MFF_ICMPV6_TYPE:
>      case MFF_ICMPV6_CODE:
> +    case MFF_TUN_GTPU_MSGTYPE:
>          return OFPUTIL_P_NONE;
>  
>      case MFF_DP_HASH:
> @@ -2250,6 +2287,9 @@ mf_set(const struct mf_field *mf,
>      case MFF_TUN_GBP_FLAGS:
>          match_set_tun_gbp_flags_masked(match, value->u8, mask->u8);
>          break;
> +    case MFF_TUN_GTPU_FLAGS:
> +        match_set_tun_gtpu_flags_masked(match, value->u8, mask->u8);
> +        break;
>      case MFF_TUN_TTL:
>          match_set_tun_ttl_masked(match, value->u8, mask->u8);
>          break;
> diff --git a/lib/meta-flow.xml b/lib/meta-flow.xml
> index 933d4b8..d169772 100644
> --- a/lib/meta-flow.xml
> +++ b/lib/meta-flow.xml
> @@ -731,6 +731,12 @@ tcp,tp_src=0x07c0/0xfff0
>        Used by Open vSwitch for NSH extensions, in the absence of an official
>        ONF-assigned class.  (This OUI is randomly generated.)
>      </dd>
> +
> +    <dt>0x005ad651 (<code>NXOXM_GTPU</code>)</dt>
> +    <dd>
> +      Used by Open vSwitch for GTP-U extensions, in the absence of an official
> +      ONF-assigned class.  (This OUI is randomly generated.)
> +    </dd>
>    </dl>
>  
>    <p>
> @@ -1899,6 +1905,68 @@ ovs-ofctl add-flow br0 tun_metadata0=1234,actions=controller
>        </p>
>      </field>
>  
> +    <h2>GTP-U Fields</h2>
> +
> +    <p>
> +      The GTP-U (v1) header is defined as follows [3GPP TS 29.281].
> +    </p>
> +
> +    <diagram>
> +      <header name="GTP-U tunnel flags">
> +        <bits name="V" above="3" width="0.15"/>
> +        <bits name="PT" above="1" width="0.15"/>
> +        <bits name="(*)" above="1" width="0.15"/>
> +        <bits name="E" above="1" width="0.15"/>
> +        <bits name="S" above="1" width="0.15"/>
> +        <bits name="PN" above="1" width="0.15"/>
> +      </header>
> +      <nospace/>
> +      <header>
> +         <bits name="Message Type" above="8" width="0.15"/>
> +         <bits name="Length" above="16" width="0.15"/>
> +         <bits name="TEID" above="32" width="0.15"/>
> +         <bits name="..."  above="" width="0.15"/>
> +      </header>
> +    </diagram>
> +
> +    <p>
> +       where
> +    </p>
> +    <ul>
> +      <li><code>V</code>: version fields, used to determine the version of the
> +      GTP-U protocol, which should be set to '1'.</li>
> +      <li><code>PT</code>: Packet Type, used as a protocol discriminator
> +      between GTP (when PT is '1') and GTP' (when PT is '0').</li>
> +      <li><code>E</code>: Extension Header flag, indicating the presence of a
> +      meaningful value of the Next Extension Header field.</li>
> +      <li><code>S</code>: Sequence number flag, indicating the presence of a
> +      meaningful value of the Sequence Number field.</li>
> +      <li><code>PN</code>: N-PDU Number flag, indicating the presence of a
> +      meaningful value of the N-PDU Number field.</li>
> +      <li><code>Message Type</code>: indicating the type of GTP-U message.
> +      </li>
> +      <li><code>Length</code>: indicating the length in octets of the payload.
> +      </li>
> +      <li><code>TEID</code>: Tunnel Endpoint Identifier,unambiguously
> +      identifying a tunnel endpoint in the receiving GTP-U protocol entity.
> +      Open vSwitch makes TEID availabe via <ref field="tun_id"/>.</li>
> +    </ul>
> +    <field id="MFF_TUN_GTPU_FLAGS" title="GTP-U tunnel flags">
> +      <p>
> +        For a packet tunneled over GTP-U, this field indicates the presence of
> +        optional headers.
> +      </p>
> +    </field>
> +
> +    <field id="MFF_TUN_GTPU_MSGTYPE" title="GTP-U tunnel message type">
> +      <p>
> +        For a packet tunneled over GTP-U, this field indicates whether it's a
> +        signalling message used for path management, or a user plane message
> +        which carries the original packet. The complete range of message types
> +        can be referred to [3GPP TS 29.281].
> +      </p>
> +    </field>
> +
>      <!-- Open vSwitch uses the following fields internally, but it
>           does not expose them to the user via OpenFlow, so we do not
>           document them. -->
> @@ -4662,6 +4730,14 @@ r r c c c.
>        <url href="https://tools.ietf.org/html/rfc7665"/>.
>      </dd>
>  
> +    <dt>3GPP TS 29.281</dt>
> +    <dd>
> +      3GPP,
> +      ``General Packet Radio System (GPRS) Tunnelling Protocol User Plane
> +      (GTPv1-U),''
> +      <url href="http://www.3gpp.org/ftp/Specs/html-info/29281.htm"/>.
> +    </dd>
> +
>      <dt>Srinivasan</dt>
>      <dd>
>        V. Srinivasan, S. Suriy, and G. Varghese, ``Packet
> diff --git a/lib/netdev-native-tnl.c b/lib/netdev-native-tnl.c
> index c3e698d..cace175 100644
> --- a/lib/netdev-native-tnl.c
> +++ b/lib/netdev-native-tnl.c
> @@ -55,6 +55,9 @@ static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(60, 5);
>  #define GENEVE_BASE_HLEN   (sizeof(struct udp_header) +         \
>                              sizeof(struct genevehdr))
>  
> +#define GTPU_HLEN   (sizeof(struct udp_header) +        \
> +                     sizeof(struct gtpuhdr))
> +
>  uint16_t tnl_udp_port_min = 32768;
>  uint16_t tnl_udp_port_max = 61000;
>  
> @@ -715,7 +718,147 @@ netdev_geneve_build_header(const struct netdev *netdev,
>      return 0;
>  }
>  
> -
> +void
> +netdev_gtpu_push_header(struct dp_packet *packet,
> +                        const struct ovs_action_push_tnl *data)
> +{
> +    struct udp_header *udp;
> +    struct gtpuhdr *gtph;
> +    int gtpu_len;
> +    int ip_tot_size;
> +
> +    gtpu_len = dp_packet_size(packet);
> +
> +    udp = netdev_tnl_push_ip_header(packet, data->header, data->header_len,
> +                                    &ip_tot_size);
> +    udp->udp_len = htons(ip_tot_size);
> +
> +    gtph = (struct gtpuhdr *)(udp + 1);
> +    gtph->len = htons(gtpu_len);
> +
> +    if (udp->udp_csum) {
> +        uint32_t csum;
> +        if (netdev_tnl_is_header_ipv6(dp_packet_data(packet))) {
> +            csum = packet_csum_pseudoheader6(
> +                       netdev_tnl_ipv6_hdr(dp_packet_data(packet)));
> +        } else {
> +            csum = packet_csum_pseudoheader(
> +                       netdev_tnl_ip_hdr(dp_packet_data(packet)));
> +        }
> +
> +        csum = csum_continue(csum, udp, ip_tot_size);
> +        udp->udp_csum = csum_finish(csum);
> +
> +        if (!udp->udp_csum) {
> +            udp->udp_csum = htons(0xffff);
> +        }
> +    }
> +
> +    packet->packet_type = htonl(PT_ETH);
> +}
> +
> +
> +int
> +netdev_gtpu_build_header(const struct netdev *netdev,
> +                          struct ovs_action_push_tnl *data,
> +                          const struct netdev_tnl_build_header_params *params)
> +{
> +    struct netdev_vport *dev = netdev_vport_cast(netdev);
> +    struct netdev_tunnel_config *tnl_cfg;
> +    struct gtpuhdr *gtph;
> +    struct udp_header *udp;
> +
> +    /* XXX: RCUfy tnl_cfg. */
> +    ovs_mutex_lock(&dev->mutex);
> +
> +    tnl_cfg = &dev->tnl_cfg;
> +    udp = netdev_tnl_ip_build_header(data, params, IPPROTO_UDP);
> +    udp->udp_dst = tnl_cfg->dst_port;
> +    udp->udp_src = htons(GTPU_DST_PORT);
> +
> +    if (params->is_ipv6 || params->flow->tunnel.flags & FLOW_TNL_F_CSUM) {
> +        /* Write a value in now to mark that we should compute the checksum
> +         * later. 0xffff is handy because it is transparent to the
> +         * calculation. */
> +        udp->udp_csum = htons(0xffff);
> +    }
> +
> +    ovs_mutex_unlock(&dev->mutex);
> +    data->header_len += sizeof *udp;
> +
> +    gtph = (struct gtpuhdr *)(udp + 1);
> +    gtph->flags = params->flow->tunnel.gtpu_flags;
> +    gtph->msgtype = params->flow->tunnel.gtpu_msgtype;
> +
> +    put_16aligned_be32(&gtph->teid,
> +                       htonl(ntohll(params->flow->tunnel.tun_id)));
> +    data->header_len += sizeof *gtph;
> +    data->tnl_type = OVS_VPORT_TYPE_GTPU;
> +    return 0;
> +}
> +
> +struct dp_packet *
> +netdev_gtpu_pop_header(struct dp_packet *packet)
> +{
> +    struct pkt_metadata *md = &packet->md;
> +    struct flow_tnl *tnl = &md->tunnel;
> +    struct gtpuhdr *gtph;
> +    unsigned int hlen;
> +
> +    pkt_metadata_init_tnl(md);
> +    if (GTPU_HLEN > dp_packet_l4_size(packet)) {
> +        VLOG_WARN_RL(&err_rl, "GTP-U packet too small: min header=%u "
> +                              "packet size=%"PRIuSIZE"\n",
> +                     (unsigned int)GTPU_HLEN, dp_packet_l4_size(packet));
> +        goto err;
> +    }
> +
> +    gtph = udp_extract_tnl_md(packet, tnl, &hlen);
> +    if (!gtph) {
> +        goto err;
> +    }
> +
> +    if (gtph->flags == 0x30) {
> +    /* Only GTP-U v1 packets without optional fileds are processed, i.e.
> +     *     8   7   6  5    4   3   2   1
> +     *    |   ver   | PT | 0 | E | S | PN |
> +     *     0   0   1  1    0   0   0   0
> +     */
> +        tnl->gtpu_flags = gtph->flags;
> +    } else {
> +        goto err;
> +    }
> +    tnl->gtpu_msgtype = gtph->msgtype;
> +    tnl->tun_id = htonll(ntohl(get_16aligned_be32(&gtph->teid)));
> +
> +    /*Figure out whether the inner packet is IPv4, IPv6 or a GTP-U message.*/
> +    if (tnl->gtpu_msgtype == GTPU_MSGTYPE_GPDU) {
> +        struct ip_header *ip;
> +
> +        ip = (struct ip_header *)(gtph + 1);
> +        if (IP_VER(ip->ip_ihl_ver) == 4) {
> +            packet->packet_type = htonl(PT_IPV4);
> +        } else if (IP_VER(ip->ip_ihl_ver) == 6) {
> +            packet->packet_type = htonl(PT_IPV6);
> +        } else {
> +            goto err;
> +        }
> +    } else {
> +        /* GTP-U messages, including echo request, end marker, etc.
> +         * OVS will be kicking it to a control-plane entity if properly
> +         * configured.
> +         */
> +        packet->packet_type = htonl(PT_GTPU_MSG);
> +    }
> +
> +    dp_packet_reset_packet(packet, hlen + GTPU_HLEN);
> +
> +    return packet;
> +err:
> +    dp_packet_delete(packet);
> +    return NULL;
> +}
> +
>  void
>  netdev_tnl_egress_port_range(struct unixctl_conn *conn, int argc,
>                               const char *argv[], void *aux OVS_UNUSED)
> diff --git a/lib/netdev-native-tnl.h b/lib/netdev-native-tnl.h
> index a912ce9..9c4cc66 100644
> --- a/lib/netdev-native-tnl.h
> +++ b/lib/netdev-native-tnl.h
> @@ -58,6 +58,18 @@ netdev_vxlan_build_header(const struct netdev *netdev,
>  struct dp_packet *
>  netdev_vxlan_pop_header(struct dp_packet *packet);
>  
> +void
> +netdev_gtpu_push_header(struct dp_packet *packet,
> +                        const struct ovs_action_push_tnl *data);
> +
> +int
> +netdev_gtpu_build_header(const struct netdev *netdev,
> +                         struct ovs_action_push_tnl *data,
> +                         const struct netdev_tnl_build_header_params *params);
> +
> +struct dp_packet *
> +netdev_gtpu_pop_header(struct dp_packet *packet);
> +
>  static inline bool
>  netdev_tnl_is_header_ipv6(const void *header)
>  {
> diff --git a/lib/netdev-vport.c b/lib/netdev-vport.c
> index 52aa12d..ee454a1 100644
> --- a/lib/netdev-vport.c
> +++ b/lib/netdev-vport.c
> @@ -107,7 +107,10 @@ netdev_vport_needs_dst_port(const struct netdev *dev)
>  
>      return (class->get_config == get_tunnel_config &&
>              (!strcmp("geneve", type) || !strcmp("vxlan", type) ||
> -             !strcmp("lisp", type) || !strcmp("stt", type)) );
> +             !strcmp("lisp", type) || !strcmp("stt", type) ||
> +             !strcmp("gtpu", type)
> +            )
> +           );
>  }
>  
>  const char *
> @@ -199,6 +202,8 @@ netdev_vport_construct(struct netdev *netdev_)
>          dev->tnl_cfg.dst_port = htons(LISP_DST_PORT);
>      } else if (!strcmp(type, "stt")) {
>          dev->tnl_cfg.dst_port = htons(STT_DST_PORT);
> +    } else if (!strcmp(type, "gtpu")) {
> +        dev->tnl_cfg.dst_port = htons(GTPU_DST_PORT);
>      }
>  
>      dev->tnl_cfg.dont_fragment = true;
> @@ -408,6 +413,8 @@ tunnel_supported_layers(const char *type,
>          return TNL_L3;
>      } else if (!strcmp(type, "gre")) {
>          return TNL_L2 | TNL_L3;
> +    } else if (!strcmp(type, "gtpu")) {
> +        return TNL_L3;
>      } else if (!strcmp(type, "vxlan")
>                 && tnl_cfg->exts & (1 << OVS_VXLAN_EXT_GPE)) {
>          return TNL_L2 | TNL_L3;
> @@ -455,6 +462,10 @@ set_tunnel_config(struct netdev *dev_, const struct smap *args, char **errp)
>          tnl_cfg.dst_port = htons(STT_DST_PORT);
>      }
>  
> +    if (!strcmp(type, "gtpu")) {
> +        tnl_cfg.dst_port = htons(GTPU_DST_PORT);
> +    }
> +
>      needs_dst_port = netdev_vport_needs_dst_port(dev_);
>      tnl_cfg.dont_fragment = true;
>  
> @@ -700,6 +711,7 @@ get_tunnel_config(const struct netdev *dev, struct smap *args)
>          if ((!strcmp("geneve", type) && dst_port != GENEVE_DST_PORT) ||
>              (!strcmp("vxlan", type) && dst_port != VXLAN_DST_PORT) ||
>              (!strcmp("lisp", type) && dst_port != LISP_DST_PORT) ||
> +            (!strcmp("gtpu", type) && dst_port != GTPU_DST_PORT) ||
>              (!strcmp("stt", type) && dst_port != STT_DST_PORT)) {
>              smap_add_format(args, "dst_port", "%d", dst_port);
>          }
> @@ -979,6 +991,10 @@ netdev_vport_tunnel_register(void)
>                                             NETDEV_VPORT_GET_IFINDEX),
>          TUNNEL_CLASS("lisp", "lisp_sys", NULL, NULL, NULL, NULL),
>          TUNNEL_CLASS("stt", "stt_sys", NULL, NULL, NULL, NULL),
> +        TUNNEL_CLASS("gtpu", "gptu_sys", netdev_gtpu_build_header,
> +                                         netdev_gtpu_push_header,
> +                                         netdev_gtpu_pop_header,
> +                                         NULL),
>      };
>      static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
>  
> diff --git a/lib/nx-match.c b/lib/nx-match.c
> index a8edb2e..c7f6c27 100644
> --- a/lib/nx-match.c
> +++ b/lib/nx-match.c
> @@ -1153,6 +1153,10 @@ nx_put_raw(struct ofpbuf *b, enum ofp_version oxm, const struct match *match,
>                  flow->tunnel.gbp_id, match->wc.masks.tunnel.gbp_id);
>      nxm_put_8m(&ctx, MFF_TUN_GBP_FLAGS, oxm,
>                 flow->tunnel.gbp_flags, match->wc.masks.tunnel.gbp_flags);
> +    nxm_put_8m(&ctx, MFF_TUN_GTPU_FLAGS, oxm, flow->tunnel.gtpu_flags,
> +               match->wc.masks.tunnel.gtpu_flags);
> +    nxm_put_8m(&ctx, MFF_TUN_GTPU_MSGTYPE, oxm, flow->tunnel.gtpu_msgtype,
> +               match->wc.masks.tunnel.gtpu_msgtype);
>      tun_metadata_to_nx_match(b, oxm, match);
>  
>      /* Network Service Header */
> diff --git a/lib/odp-util.c b/lib/odp-util.c
> index 70188b6..2c5a1f1 100644
> --- a/lib/odp-util.c
> +++ b/lib/odp-util.c
> @@ -712,6 +712,15 @@ format_odp_tnl_push_header(struct ds *ds, struct ovs_action_push_tnl *data)
>              options++;
>          }
>          ds_put_format(ds, ")");
> +    } else if (data->tnl_type == OVS_VPORT_TYPE_GTPU) {
> +        const struct gtpuhdr *gtph;
> +
> +        gtph = format_udp_tnl_push_header(ds, udp);
> +
> +        ds_put_format(ds, "gtpu(flags=0x%"PRIx8
> +                          ",msgtype=0x%"PRIx8",teid=0x%"PRIx32")",
> +                      gtph->flags, gtph->msgtype,
> +                      ntohl(get_16aligned_be32(&gtph->teid)));
>      }
>      ds_put_format(ds, ")");
>  }
> @@ -1474,6 +1483,7 @@ ovs_parse_tnl_push(const char *s, struct ovs_action_push_tnl *data)
>      if (ovs_scan_len(s, &n, "udp(src=%"SCNi16",dst=%"SCNi16",csum=0x%"SCNx16"),",
>                       &udp_src, &udp_dst, &csum)) {
>          uint32_t vx_flags, vni;
> +        uint8_t gtpu_flags, gtpu_msgtype;
>  
>          udp->udp_src = htons(udp_src);
>          udp->udp_dst = htons(udp_dst);
> @@ -1528,6 +1538,17 @@ ovs_parse_tnl_push(const char *s, struct ovs_action_push_tnl *data)
>              gnh->proto_type = htons(ETH_TYPE_TEB);
>              put_16aligned_be32(&gnh->vni, htonl(vni << 8));
>              tnl_type = OVS_VPORT_TYPE_GENEVE;
> +        } else if (ovs_scan_len(s, &n, "gtpu(flags=0x%"SCNx8",msgtype=0x%"
> +                                       SCNx8",teid=0x%"SCNx32"))",
> +                                &gtpu_flags, &gtpu_msgtype, &vni)) {
> +            struct gtpuhdr *gtph = (struct gtpuhdr *) (udp + 1);
> +
> +            gtph->flags = gtpu_flags;
> +            gtph->msgtype = gtpu_msgtype;
> +            put_16aligned_be32(&gtph->teid, htonl(vni));
> +            tnl_type = OVS_VPORT_TYPE_GTPU;
> +            header_len = sizeof *eth + ip_len +
> +                         sizeof *udp + sizeof *gtph;
>          } else {
>              return -EINVAL;
>          }
> @@ -2373,6 +2394,8 @@ static const struct attr_len_tbl ovs_tun_key_attr_lens[OVS_TUNNEL_KEY_ATTR_MAX +
>      [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = ATTR_LEN_NESTED,
>                                              .next = ovs_vxlan_ext_attr_lens ,
>                                              .next_max = OVS_VXLAN_EXT_MAX},
> +    [OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS]    = { .len = 1 },
> +    [OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE]  = { .len = 1 },
>      [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = 16 },
>      [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = 16 },
>  };
> @@ -2698,6 +2721,12 @@ odp_tun_key_from_attr__(const struct nlattr *attr, bool is_mask,
>          case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
>              tun_metadata_from_geneve_nlattr(a, is_mask, tun);
>              break;
> +        case OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS:
> +            tun->gtpu_flags = nl_attr_get_u8(a);
> +            break;
> +        case OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE:
> +            tun->gtpu_msgtype = nl_attr_get_u8(a);
> +            break;
>  
>          default:
>              /* Allow this to show up as unexpected, if there are unknown
> @@ -2767,6 +2796,13 @@ tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl *tun_key,
>      if (tun_key->flags & FLOW_TNL_F_OAM) {
>          nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_OAM);
>      }
> +    if (tun_key->gtpu_flags) {
> +        nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS, tun_key->gtpu_flags);
> +    }
> +    if (tun_key->gtpu_msgtype) {
> +        nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE,
> +                      tun_key->gtpu_msgtype);
> +    }
>  
>      /* If tnl_type is set to a particular type of output tunnel,
>       * only put its relevant tunnel metadata to the nlattr.
> @@ -3487,6 +3523,14 @@ format_odp_tun_attr(const struct nlattr *attr, const struct nlattr *mask_attr,
>              format_odp_tun_geneve(a, ma, ds, verbose);
>              ds_put_cstr(ds, "),");
>              break;
> +        case OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS:
> +            format_u8x(ds, "gtpu_flags", nl_attr_get_u8(a),
> +                       ma ? nl_attr_get(ma) : NULL, verbose);
> +            break;
> +        case OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE:
> +            format_u8x(ds, "gtpu_msgtype", nl_attr_get_u8(a),
> +                       ma ? nl_attr_get(ma) : NULL, verbose);
> +            break;
>          case OVS_TUNNEL_KEY_ATTR_PAD:
>              break;
>          case __OVS_TUNNEL_KEY_ATTR_MAX:
> @@ -5179,6 +5223,10 @@ parse_odp_key_mask_attr(const char *s, const struct simap *port_names,
>          SCAN_FIELD_NESTED_FUNC("geneve(", struct geneve_scan, geneve,
>                                 geneve_to_attr);
>          SCAN_FIELD_NESTED_FUNC("flags(", uint16_t, tun_flags, tun_flags_to_attr);
> +        SCAN_FIELD_NESTED("gtpu_flags=", uint8_t, u8,
> +                          OVS_TUNNEL_KEY_ATTR_GTPU_FLAGS);
> +        SCAN_FIELD_NESTED("gtpu_msg_type=", uint8_t, u8,
> +                          OVS_TUNNEL_KEY_ATTR_GTPU_MSGTYPE);
>      } SCAN_END_NESTED();
>  
>      SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT);
> diff --git a/lib/packets.h b/lib/packets.h
> index b2bf706..16cb3f3 100644
> --- a/lib/packets.h
> +++ b/lib/packets.h
> @@ -1306,6 +1306,23 @@ BUILD_ASSERT_DECL(sizeof(struct vxlanhdr) == 8);
>  #define VXLAN_F_GPE  0x4000
>  #define VXLAN_HF_GPE 0x04000000
>  
> +/* GTP-U protocol header */
> +struct gtpuhdr {
> +    uint8_t flags;
> +    uint8_t msgtype;
> +    ovs_be16 len;
> +    ovs_16aligned_be32 teid;
> +};
> +
> +/* GTP-U UDP port */
> +#define GTPU_DST_PORT 2152
> +
> +/* Default GTP-U flags */
> +#define GTPU_FLAGS_DEFAULT 0x30
> +
> +/* GTP-U message type for normal user plane PDU */
> +#define GTPU_MSGTYPE_GPDU 255
> +
>  /* Input values for PACKET_TYPE macros have to be in host byte order.
>   * The _BE postfix indicates result is in network byte order. Otherwise result
>   * is in host byte order. */
> @@ -1342,6 +1359,8 @@ enum packet_type {
>      PT_MPLS = PACKET_TYPE(OFPHTN_ETHERTYPE, ETH_TYPE_MPLS),
>      PT_MPLS_MC = PACKET_TYPE(OFPHTN_ETHERTYPE, ETH_TYPE_MPLS_MCAST),
>      PT_NSH  = PACKET_TYPE(OFPHTN_ETHERTYPE, ETH_TYPE_NSH),
> +    /* GTP-U message. */
> +    PT_GTPU_MSG = PACKET_TYPE(OFPHTN_UDP_TCP_PORT, GTPU_DST_PORT),
>      PT_UNKNOWN = PACKET_TYPE(0xffff, 0xffff),   /* Unknown packet type. */
>  };
>  
> diff --git a/lib/tnl-ports.c b/lib/tnl-ports.c
> index b814f7a..cb8fab8 100644
> --- a/lib/tnl-ports.c
> +++ b/lib/tnl-ports.c
> @@ -177,6 +177,9 @@ tnl_type_to_nw_proto(const char type[])
>      if (!strcmp(type, "vxlan")) {
>          return IPPROTO_UDP;
>      }
> +    if (!strcmp(type, "gtpu")) {
> +        return IPPROTO_UDP;
> +    }
>      return 0;
>  }
>  
> diff --git a/ofproto/ofproto-dpif-ipfix.c b/ofproto/ofproto-dpif-ipfix.c
> index 4d9fe78..2951a74 100644
> --- a/ofproto/ofproto-dpif-ipfix.c
> +++ b/ofproto/ofproto-dpif-ipfix.c
> @@ -88,6 +88,7 @@ enum dpif_ipfix_tunnel_type {
>      DPIF_IPFIX_TUNNEL_LISP = 0x03,
>      DPIF_IPFIX_TUNNEL_STT = 0x04,
>      DPIF_IPFIX_TUNNEL_GENEVE = 0x07,
> +    DPIF_IPFIX_TUNNEL_GTPU = 0x08,
>      NUM_DPIF_IPFIX_TUNNEL
>  };
>  
> @@ -389,6 +390,7 @@ static uint8_t tunnel_protocol[NUM_DPIF_IPFIX_TUNNEL] = {
>      IPPROTO_TCP,    /* DPIF_IPFIX_TUNNEL_STT*/
>      0          ,    /* reserved */
>      IPPROTO_UDP,    /* DPIF_IPFIX_TUNNEL_GENEVE*/
> +    IPPROTO_UDP,    /* DPIF_IPFIX_TUNNEL_GTPU */
>  };
>  
>  OVS_PACKED(
> @@ -509,6 +511,7 @@ BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_aggregated_tcp) == 48);
>   * support tunnel key for:
>   * VxLAN: 24-bit VIN,
>   * GRE: 32-bit key,
> + * GTPU: 32-bit key,
>   * LISP: 24-bit instance ID
>   * STT: 64-bit key
>   */
> @@ -811,6 +814,8 @@ dpif_ipfix_tunnel_type(const struct ofport *ofport)
>          return DPIF_IPFIX_TUNNEL_GENEVE;
>      } else if (strcmp(type, "stt") == 0) {
>          return DPIF_IPFIX_TUNNEL_STT;
> +    } else if (strcmp(type, "gtpu") == 0) {
> +        return DPIF_IPFIX_TUNNEL_GTPU;
>      }
>  
>      return DPIF_IPFIX_TUNNEL_UNKNOWN;
> diff --git a/ofproto/ofproto-dpif-sflow.c b/ofproto/ofproto-dpif-sflow.c
> index bbec30e..042c7cf 100644
> --- a/ofproto/ofproto-dpif-sflow.c
> +++ b/ofproto/ofproto-dpif-sflow.c
> @@ -61,7 +61,8 @@ enum dpif_sflow_tunnel_type {
>      DPIF_SFLOW_TUNNEL_VXLAN,
>      DPIF_SFLOW_TUNNEL_GRE,
>      DPIF_SFLOW_TUNNEL_LISP,
> -    DPIF_SFLOW_TUNNEL_GENEVE
> +    DPIF_SFLOW_TUNNEL_GENEVE,
> +    DPIF_SFLOW_TUNNEL_GTPU
>  };
>  
>  struct dpif_sflow_port {
> @@ -621,6 +622,8 @@ dpif_sflow_tunnel_type(struct ofport *ofport) {
>  	    return DPIF_SFLOW_TUNNEL_VXLAN;
>  	} else if (strcmp(type, "lisp") == 0) {
>  	    return DPIF_SFLOW_TUNNEL_LISP;
> +	} else if (strcmp(type, "gtpu") == 0) {
> +	    return DPIF_SFLOW_TUNNEL_GTPU;
>  	} else if (strcmp(type, "geneve") == 0) {
>  	    return DPIF_SFLOW_TUNNEL_GENEVE;
>  	}
> @@ -641,6 +644,7 @@ dpif_sflow_tunnel_proto(enum dpif_sflow_tunnel_type tunnel_type)
>  
>      case DPIF_SFLOW_TUNNEL_VXLAN:
>      case DPIF_SFLOW_TUNNEL_LISP:
> +    case DPIF_SFLOW_TUNNEL_GTPU:
>      case DPIF_SFLOW_TUNNEL_GENEVE:
>          ipproto = IPPROTO_UDP;
>  
> diff --git a/ofproto/ofproto-dpif-xlate.c b/ofproto/ofproto-dpif-xlate.c
> index 3cb7b5e..f566e3c 100644
> --- a/ofproto/ofproto-dpif-xlate.c
> +++ b/ofproto/ofproto-dpif-xlate.c
> @@ -3352,6 +3352,7 @@ propagate_tunnel_data_to_flow(struct xlate_ctx *ctx, struct eth_addr dmac,
>          break;
>      case OVS_VPORT_TYPE_VXLAN:
>      case OVS_VPORT_TYPE_GENEVE:
> +    case OVS_VPORT_TYPE_GTPU:
>          nw_proto = IPPROTO_UDP;
>          break;
>      case OVS_VPORT_TYPE_LISP:
> @@ -7166,7 +7167,8 @@ xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
>          ctx.xin->xport_uuid = in_port->uuid;
>      }
>  
> -    if (flow->packet_type != htonl(PT_ETH) && in_port &&
> +    if (flow->packet_type != htonl(PT_GTPU_MSG) &&
> +        flow->packet_type != htonl(PT_ETH) && in_port &&
>          in_port->pt_mode == NETDEV_PT_LEGACY_L3 && ctx.table_id == 0) {
>          /* Add dummy Ethernet header to non-L2 packet if it's coming from a
>           * L3 port. So all packets will be L2 packets for lookup.
> diff --git a/ofproto/tunnel.c b/ofproto/tunnel.c
> index a040b54..7de68c4 100644
> --- a/ofproto/tunnel.c
> +++ b/ofproto/tunnel.c
> @@ -384,6 +384,12 @@ tnl_wc_init(struct flow *flow, struct flow_wildcards *wc)
>          wc->masks.tunnel.tp_src = 0;
>          wc->masks.tunnel.tp_dst = 0;
>  
> +        /* GTP-U header are set to be always unwildcarded for GTP-U packets*/
> +        if (flow->tunnel.gtpu_flags || flow->tunnel.gtpu_msgtype) {
> +            wc->masks.tunnel.gtpu_flags = UINT8_MAX;
> +            wc->masks.tunnel.gtpu_msgtype = UINT8_MAX;
> +        }
> +
>          if (is_ip_any(flow)
>              && IP_ECN_is_ce(flow->tunnel.ip_tos)) {
>              wc->masks.nw_tos |= IP_ECN_MASK;
> @@ -569,7 +575,8 @@ tnl_find(const struct flow *flow) OVS_REQ_RDLOCK(rwlock)
>                      match.ip_src_flow = ip_src == IP_SRC_FLOW;
>  
>                      /* Look for a legacy L2 or L3 tunnel port first. */
> -                    if (pt_ns(flow->packet_type) == OFPHTN_ETHERTYPE) {
> +                    if (pt_ns(flow->packet_type) == OFPHTN_ETHERTYPE ||
> +                        flow->packet_type == htonl(PT_GTPU_MSG)) {
>                          match.pt_mode = NETDEV_PT_LEGACY_L3;
>                      } else {
>                          match.pt_mode = NETDEV_PT_LEGACY_L2;
> diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
> index 7ab90d5..83f6cbb 100644
> --- a/vswitchd/vswitch.xml
> +++ b/vswitchd/vswitch.xml
> @@ -2283,6 +2283,22 @@
>              </p>
>            </dd>
>  
> +          <dt><code>gtpu</code></dt>
> +          <dd>
> +              GPRS Tunneling Protocol (GTP) is a group of IP-based communications
> +              protocols used to carry general packet radio service (GPRS) within GSM,
> +              UMTS and LTE networks. GTP-U is used for carrying user data within the GPRS
> +              core network and between the radio access network and the core network.
> +              The user data transported can be packets in any of IPv4, IPv6, or PPP
> +              formats.
> +
> +              The protocol is documented at http://www.3gpp.org/DynaReport/29281.htm
> +
> +              Open vSwitch uses UDP destination port 2152. The source port used for
> +              GTP traffic varies on a per-flow basis and is in the ephemeral port
> +              range.
> +          </dd>
> +
>            <dt><code>stt</code></dt>
>            <dd>
>              The Stateless TCP Tunnel (STT) is particularly useful when tunnel

On Thu, May 17, 2018 at 05:11:24AM +0800, Aaron Conole wrote:
> Yi Yang <yi.y.yang@intel.com> writes:
> 
> Hi Yi,
> 
> Thanks for the patch!  Just a brief review.
> 

Aaron, thank you so much for your quick review.

> >
> > Signed-off-by: Feng Yang <feng.yang@intel.com>
> > Signed-off-by: Jiannan Ouyang <ouyangj@fb.com>
> > Signed-off-by: Yi Yang <yi.y.yang@intel.com>
> 
> Needs Co-author tags.

Ok, will add it in v3.

> >  OXM_CLASSES = {"NXM_OF_":        (0,          0x0000),
> >                 "NXM_NX_":        (0,          0x0001),
> >                 "NXOXM_NSH_":     (0x005ad650, 0xffff),
> > +               "NXOXM_GTPU_":    (0x005ad651, 0xffff),
> >                 "OXM_OF_":        (0,          0x8000),
> >                 "OXM_OF_PKT_REG": (0,          0x8001),
> >                 "ONFOXM_ET_":     (0x4f4e4600, 0xffff),
> > diff --git a/datapath/linux/compat/include/linux/openvswitch.h b/datapath/linux/compat/include/linux/openvswitch.h
> > index 84ebcaf..ad6ea64 100644
> > --- a/datapath/linux/compat/include/linux/openvswitch.h
> > +++ b/datapath/linux/compat/include/linux/openvswitch.h
> 
> I don't see this in the upstream code.  This is introducing another
> place for kernel and ovs to diverge.  Please get this type accepted
> upstream first.

You're partially right :-), there was the case OVS accepted first, then
kernel did so. I think OVS community is very open for this.

> > diff --git a/lib/dpif-netlink-rtnl.c b/lib/dpif-netlink-rtnl.c
> > index 40c4569..a628b72 100644
> > --- a/lib/dpif-netlink-rtnl.c
> > +++ b/lib/dpif-netlink-rtnl.c
> 
> Again, you'll need to get the netlink datapath changes through kernel
> process.  At least try.

Yeah, I will, because Facebook guy Jiannan Ouyang did that before, they
won't push this, so I'll try pushing this, but you know this is a slow
process.

> >          break;
> > @@ -814,6 +817,8 @@ netdev_to_ovs_vport_type(const char *type)
> >          return OVS_VPORT_TYPE_VXLAN;
> >      } else if (!strcmp(type, "lisp")) {
> >          return OVS_VPORT_TYPE_LISP;
> > +    } else if (strstr(type, "gtpu")) {
> 
> Why isn't this using "!strcmp"?

good catch, it should be "!strcmp", I'll change it in v3.

> >  
> > +    /* It won't be parsed if the packet contains application layer data only.
> > +     * These include, but not limited to GTP-U messages, GTP-C packets.
> > +     */
> 
> Why treat GTP-U and C separate here?

GTP-C is very complicated and it is for control plane, so it won't be
helpful to have it in OVS datapath, Joe mentioned this in v1 comments, we discussed
this, GTP-U is what we care in real user scenarios.

In addition, GPT-C and GTP-U use different UDP port, GTP-U tunnel can't
handle GTP-C, here the comment may be misleading you. It means ovs won't
parse the packet if it is GTP-U signaling message but not normal GTP-U
PDU.

Hi Aaron, Yi,
Good questions and explanation.
Regarding the GTP-U and GTP-C issue, the comment that "These include, but not limited to GTP-U messages, GTP-C packets" sounds misleading indeed.
We will remove "GTP-C packets" in v3.  

Regards, Feng
2018-5-17
-----Original Message-----
From: Yang, Yi Y 
Sent: Thursday, May 17, 2018 10:22 AM
To: Aaron Conole <aconole@redhat.com>
Cc: dev@openvswitch.org; Yang, Feng <feng.yang@intel.com>
Subject: Re: [ovs-dev] [PATCH v2 1/2] userspace datapath: Add GTP-U tunnel support

On Thu, May 17, 2018 at 05:11:24AM +0800, Aaron Conole wrote:
> Yi Yang <yi.y.yang@intel.com> writes:
> 
> Hi Yi,
> 
> Thanks for the patch!  Just a brief review.
> 

Aaron, thank you so much for your quick review.

> >
> > Signed-off-by: Feng Yang <feng.yang@intel.com>
> > Signed-off-by: Jiannan Ouyang <ouyangj@fb.com>
> > Signed-off-by: Yi Yang <yi.y.yang@intel.com>
> 
> Needs Co-author tags.

Ok, will add it in v3.

> >  OXM_CLASSES = {"NXM_OF_":        (0,          0x0000),
> >                 "NXM_NX_":        (0,          0x0001),
> >                 "NXOXM_NSH_":     (0x005ad650, 0xffff),
> > +               "NXOXM_GTPU_":    (0x005ad651, 0xffff),
> >                 "OXM_OF_":        (0,          0x8000),
> >                 "OXM_OF_PKT_REG": (0,          0x8001),
> >                 "ONFOXM_ET_":     (0x4f4e4600, 0xffff),
> > diff --git a/datapath/linux/compat/include/linux/openvswitch.h 
> > b/datapath/linux/compat/include/linux/openvswitch.h
> > index 84ebcaf..ad6ea64 100644
> > --- a/datapath/linux/compat/include/linux/openvswitch.h
> > +++ b/datapath/linux/compat/include/linux/openvswitch.h
> 
> I don't see this in the upstream code.  This is introducing another 
> place for kernel and ovs to diverge.  Please get this type accepted 
> upstream first.

You're partially right :-), there was the case OVS accepted first, then kernel did so. I think OVS community is very open for this.

> > diff --git a/lib/dpif-netlink-rtnl.c b/lib/dpif-netlink-rtnl.c index 
> > 40c4569..a628b72 100644
> > --- a/lib/dpif-netlink-rtnl.c
> > +++ b/lib/dpif-netlink-rtnl.c
> 
> Again, you'll need to get the netlink datapath changes through kernel 
> process.  At least try.

Yeah, I will, because Facebook guy Jiannan Ouyang did that before, they won't push this, so I'll try pushing this, but you know this is a slow process.

> >          break;
> > @@ -814,6 +817,8 @@ netdev_to_ovs_vport_type(const char *type)
> >          return OVS_VPORT_TYPE_VXLAN;
> >      } else if (!strcmp(type, "lisp")) {
> >          return OVS_VPORT_TYPE_LISP;
> > +    } else if (strstr(type, "gtpu")) {
> 
> Why isn't this using "!strcmp"?

good catch, it should be "!strcmp", I'll change it in v3.

> >  
> > +    /* It won't be parsed if the packet contains application layer data only.
> > +     * These include, but not limited to GTP-U messages, GTP-C packets.
> > +     */
> 
> Why treat GTP-U and C separate here?

GTP-C is very complicated and it is for control plane, so it won't be helpful to have it in OVS datapath, Joe mentioned this in v1 comments, we discussed this, GTP-U is what we care in real user scenarios.

In addition, GPT-C and GTP-U use different UDP port, GTP-U tunnel can't handle GTP-C, here the comment may be misleading you. It means ovs won't parse the packet if it is GTP-U signaling message but not normal GTP-U PDU.