path: root/common/src/packet.cpp

#include <string.h>
#include <stdlib.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#define __FAVOR_BSD 1
#include <netinet/tcp.h>
#include <netinet/udp.h>
#include <netinet/ether.h>
#include <linux/ppp_defs.h>

#include "uthash.h"
#include "packet.h"

#define likely(expr) __builtin_expect((expr), 1)
#define unlikely(expr) __builtin_expect((expr), 0)

#define LOG_PACKET "PACKET"
#define PACKET_LOG_DATA_INSUFFICIENCY(type)                        \
    {                                                              \
        PACKET_LOG_ERROR("%s: layer: %s, data insufficiency",      \
                         LOG_PACKET, layer_type_tostring((type))); \
    }

#define PACKET_LOG_UNSUPPORT_PROTO(tag, next_proto)         \
    {                                                       \
        PACKET_LOG_ERROR("%s: %s: unsupport next proto %d", \
                         LOG_PACKET, (tag), (next_proto));  \
    }

/******************************************************************************
 * Static API
 ******************************************************************************/

static const char *ethproto_tostring(uint16_t proto);
static const char *ipproto_tostring(uint16_t proto);
static inline const char *ldbc_method_tostring(enum ldbc_method method);
static inline const char *layer_type_tostring(enum layer_type type);

static inline uint16_t gtp_header_get_length(const char *data, uint16_t len);
static inline uint16_t gre_header_get_length(const char *data, uint16_t len);

static inline struct layer_record *get_layer(struct packet *handler);

static inline void set_four_tuple(const char *data, enum layer_type type, struct four_tuple *tuple);
static inline void set_two_tuple(const char *data, enum layer_type type, struct two_tuple *tuple);

// 数据链路层
static inline const char *parse_ether(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_ppp(struct packet *handler, const char *data, uint16_t len);
// 数据链路层 -- 隧道
static inline const char *parse_vlan(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_pppoe_ses(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_mpls(struct packet *handler, const char *data, uint16_t len);
// 网络层
static inline const char *parse_ipv4(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_ipv6(struct packet *handler, const char *data, uint16_t len);
// 网络层 -- 隧道
static inline const char *parse_gre(struct packet *handler, const char *data, uint16_t len);
// 传输层
static inline const char *parse_udp(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_tcp(struct packet *handler, const char *data, uint16_t len);
// 传输层 -- 隧道
static inline const char *parse_vxlan(struct packet *handler, const char *data, uint16_t len);
static inline const char *parse_gtpv1_u(struct packet *handler, const char *data, uint16_t len);
// L3/L4
static inline const char *parse_l3(struct packet *handler, uint16_t next_proto, const char *data, uint16_t len);
static inline const char *parse_l4(struct packet *handler, uint8_t next_proto, const char *data, uint16_t len);

/******************************************************************************
 * Private API -- Utils
 ******************************************************************************/

static const char *ethproto_tostring(uint16_t proto)
{
    switch (proto)
    {
    case ETH_P_LOOP:
        return "LOOP";
    case ETH_P_PUP:
        return "PUP";
    case ETH_P_PUPAT:
        return "PUPAT";
    case ETH_P_IP:
        return "IP";
    case ETH_P_X25:
        return "X25";
    case ETH_P_ARP:
        return "ARP";
    case ETH_P_BPQ:
        return "BPQ";
    case ETH_P_IEEEPUP:
        return "IEEEPUP";
    case ETH_P_IEEEPUPAT:
        return "IEEEPUPAT";
    case ETH_P_DEC:
        return "DEC";
    case ETH_P_DNA_DL:
        return "DNA_DL";
    case ETH_P_DNA_RC:
        return "DNA_RC";
    case ETH_P_DNA_RT:
        return "DNA_RT";
    case ETH_P_LAT:
        return "LAT";
    case ETH_P_DIAG:
        return "DIAG";
    case ETH_P_CUST:
        return "CUST";
    case ETH_P_SCA:
        return "SCA";
    case ETH_P_TEB:
        return "TEB";
    case ETH_P_RARP:
        return "RARP";
    case ETH_P_ATALK:
        return "ATALK";
    case ETH_P_AARP:
        return "AARP";
    case ETH_P_8021Q:
        return "8021Q";
    case ETH_P_IPX:
        return "IPX";
    case ETH_P_IPV6:
        return "IPV6";
    case ETH_P_PAUSE:
        return "PAUSE";
    case ETH_P_SLOW:
        return "SLOW";
    case ETH_P_WCCP:
        return "WCCP";
    case ETH_P_PPP_DISC:
        return "PPP_DISC";
    case ETH_P_PPP_SES:
        return "PPP_SES";
    case ETH_P_MPLS_UC:
        return "MPLS_UC";
    case ETH_P_MPLS_MC:
        return "MPLS_MC";
    case ETH_P_ATMMPOA:
        return "ATMMPOA";
    case ETH_P_LINK_CTL:
        return "LINK_CTL";
    case ETH_P_ATMFATE:
        return "ATMFATE";
    case ETH_P_PAE:
        return "PAE";
    case ETH_P_AOE:
        return "AOE";
    case ETH_P_8021AD:
        return "8021AD";
    case ETH_P_802_EX1:
        return "802_EX1";
    case ETH_P_TIPC:
        return "TIPC";
    case ETH_P_8021AH:
        return "8021AH";
    case ETH_P_1588:
        return "1588";
    case ETH_P_FCOE:
        return "FCOE";
    case ETH_P_TDLS:
        return "TDLS";
    case ETH_P_FIP:
        return "FIP";
    case ETH_P_QINQ1:
        return "QINQ1";
    case ETH_P_QINQ2:
        return "QINQ2";
    case ETH_P_QINQ3:
        return "QINQ3";
    case ETH_P_EDSA:
        return "EDSA";
    case ETH_P_AF_IUCV:
        return "AF_IUCV";
    default:
        return "UNKNOWN";
    }
}

static const char *ipproto_tostring(uint16_t proto)
{
    switch (proto)
    {
    case IPPROTO_IP:
        return "IP";
    case IPPROTO_ICMP:
        return "ICMP";
    case IPPROTO_IGMP:
        return "IGMP";
    case IPPROTO_IPIP:
        return "IPIP";
    case IPPROTO_TCP:
        return "TCP";
    case IPPROTO_EGP:
        return "EGP";
    case IPPROTO_PUP:
        return "PUP";
    case IPPROTO_UDP:
        return "UDP";
    case IPPROTO_IDP:
        return "IDP";
    case IPPROTO_TP:
        return "TP";
    case IPPROTO_DCCP:
        return "DCCP";
    case IPPROTO_IPV6:
        return "IPV6";
    case IPPROTO_ROUTING:
        return "ROUTING";
    case IPPROTO_FRAGMENT:
        return "FRAGMENT";
    case IPPROTO_RSVP:
        return "RSVP";
    case IPPROTO_GRE:
        return "GRE";
    case IPPROTO_ESP:
        return "ESP";
    case IPPROTO_AH:
        return "AH";
    case IPPROTO_ICMPV6:
        return "ICMPV6";
    case IPPROTO_NONE:
        return "NONE";
    case IPPROTO_DSTOPTS:
        return "DSTOPTS";
    case IPPROTO_MTP:
        return "MTP";
    case IPPROTO_ENCAP:
        return "ENCAP";
    case IPPROTO_PIM:
        return "PIM";
    case IPPROTO_COMP:
        return "COMP";
    case IPPROTO_SCTP:
        return "SCTP";
    case IPPROTO_UDPLITE:
        return "UDPLITE";
    case IPPROTO_RAW:
        return "RAW";
    default:
        return "UNKNOWN";
    }
}

static inline const char *ldbc_method_tostring(enum ldbc_method method)
{
    switch (method)
    {
    case LDBC_METHOD_HASH_INT_IP:
        return "outter_internal_ip";
    case LDBC_METHOD_HASH_EXT_IP:
        return "outter_external_ip";
    case LDBC_METHOD_HASH_INT_IP_AND_EXT_IP:
        return "outter_internal_ip_and_external_ip";
    case LDBC_METHOD_HASH_INNERMOST_INT_IP:
        return "inner_internal_ip";
    case LDBC_METHOD_HASH_INNERMOST_EXT_IP:
        return "inner_external_ip";
    default:
        return "unknown";
    }
}

static inline const char *layer_type_tostring(enum layer_type type)
{
    switch (type)
    {
    case LAYER_TYPE_ETHER:
        return "ETH";
    case LAYER_TYPE_PPP:
        return "PPP";
    case LAYER_TYPE_HDLC:
        return "HDLC";
    case LAYER_TYPE_VLAN:
        return "VLAN";
    case LAYER_TYPE_PPPOE:
        return "PPPOE";
    case LAYER_TYPE_MPLS:
        return "MPLS";
    case LAYER_TYPE_IPV4:
        return "IPV4";
    case LAYER_TYPE_IPV6:
        return "IPV6";
    case LAYER_TYPE_GRE:
        return "GRE";
    case LAYER_TYPE_UDP:
        return "UDP";
    case LAYER_TYPE_TCP:
        return "TCP";
    case LAYER_TYPE_VXLAN:
        return "VXLAN";
    case LAYER_TYPE_GTPV1_U:
        return "GTPV1";
    default:
        return "UNKNOWN";
    }
}

static inline struct layer_record *get_layer(struct packet *handler)
{
    if (handler->layers_used >= handler->layers_size)
    {
        return NULL;
    }

    return &handler->layers[handler->layers_used];
}

#define SET_LAYER(_handler, _layer, _type, _hdr_len, _data, _len) \
    {                                                             \
        (_layer)->type = (_type);                                 \
        (_layer)->hdr_offset = (_handler)->data_len - (_len);     \
        (_layer)->hdr_ptr = (_data);                              \
        (_layer)->hdr_len = (_hdr_len);                           \
        (_layer)->pld_ptr = (_data) + (_hdr_len);                 \
        (_layer)->pld_len = (_len) - (_hdr_len);                  \
        (_handler)->layers_used++;                                \
    }

static inline void set_two_tuple(const char *data, enum layer_type type, struct two_tuple *tuple)
{
    const struct ip *ipv4 = NULL;
    const struct ip6_hdr *ipv6 = NULL;

    switch (type)
    {
    case LAYER_TYPE_IPV4:
        ipv4 = (const struct ip *)data;
        two_tuple_set_ipv4_address(tuple, &ipv4->ip_src, &ipv4->ip_dst);
        break;
    case LAYER_TYPE_IPV6:
        ipv6 = (const struct ip6_hdr *)data;
        two_tuple_set_ipv6_address(tuple, &ipv6->ip6_src, &ipv6->ip6_dst);
        break;
    default:
        break;
    }
}

static inline void set_four_tuple(const char *data, enum layer_type type, struct four_tuple *tuple)
{
    const struct ip *ipv4 = NULL;
    const struct ip6_hdr *ipv6 = NULL;
    const struct tcphdr *tcp = NULL;
    const struct udphdr *udp = NULL;

    switch (type)
    {
    case LAYER_TYPE_TCP:
        tcp = (const struct tcphdr *)data;
        four_tuple_set_port(tuple, tcp->th_sport, tcp->th_dport);
        break;
    case LAYER_TYPE_UDP:
        udp = (const struct udphdr *)data;
        four_tuple_set_port(tuple, udp->uh_sport, udp->uh_dport);
        break;
    case LAYER_TYPE_IPV4:
        ipv4 = (const struct ip *)data;
        four_tuple_set_ipv4_address(tuple, &ipv4->ip_src, &ipv4->ip_dst);
        break;
    case LAYER_TYPE_IPV6:
        ipv6 = (const struct ip6_hdr *)data;
        four_tuple_set_ipv6_address(tuple, &ipv6->ip6_src, &ipv6->ip6_dst);
        break;
    default:
        break;
    }
}

/******************************************************************************
 * Private API -- Parses
 ******************************************************************************/

static inline uint16_t gtp_header_get_length(const char *data, uint16_t len)
{
#define GTP_HDR_VER (0xE0)
#define GTP_HDR_FLAG_N_PDU (0x01)
#define GTP_HDR_FLAG_SEQ_NUM (0x02)
#define GTP_HDR_FLAG_EXT_HDR (0x04)

    struct gtp_hdr
    {
        uint8_t flags;
        uint8_t msg_type;
        uint16_t msg_len;
        uint32_t teid;
    } __attribute__((__packed__));

    struct gtp_opt
    {
        uint16_t seq_num;
        uint8_t npdu;
        uint8_t next_ext_hdr;
    } __attribute__((__packed__));

    uint16_t hdr_offset = 0;
    if (len < sizeof(struct gtp_hdr))
    {
        return 0;
    }
    const struct gtp_hdr *gtp = (const struct gtp_hdr *)data;
    hdr_offset += sizeof(struct gtp_hdr); // skip gre hdr

    // GTPv0 Not Supported
    if (((gtp->flags & GTP_HDR_VER) >> 5) != 1)
    {
        return 0;
    }

    if (gtp->flags & (GTP_HDR_FLAG_SEQ_NUM | GTP_HDR_FLAG_N_PDU | GTP_HDR_FLAG_EXT_HDR))
    {
        if (hdr_offset + sizeof(struct gtp_opt) > len)
        {
            return 0;
        }
        struct gtp_opt *opt_hdr = (struct gtp_opt *)((char *)data + hdr_offset);
        uint8_t next_ext_hdr = opt_hdr->next_ext_hdr;
        hdr_offset += sizeof(struct gtp_opt); // skip gre opt

        while (next_ext_hdr)
        {
            if (hdr_offset + 1 > len)
            {
                return 0;
            }
            uint8_t length = *((char *)data + hdr_offset) * 4 - 2;
            hdr_offset += 1; // skip length field

            if (hdr_offset + length + 1 > len)
            {
                return 0;
            }
            hdr_offset += length; // skip data field
            next_ext_hdr = *((char *)data + hdr_offset);
            hdr_offset += 1; // skip next ext hdr field
        }
    }

    return hdr_offset;
}

static inline uint16_t gre_header_get_length(const char *data, uint16_t len)
{
    /*
     * GRE Header Format (Version 0)
     *
     *   0                   1                   2                   3
     *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |C|R|K|S|s|Recur|  Flags  | Ver |         Protocol Type         |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |      Checksum (optional)      |       Offset (optional)       |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |                         Key (optional)                        |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |                    Sequence Number (optional)                 |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |                         Routing (optional)
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *
     *   0                   1                   2                   3
     *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |       Address Family          |  SRE Offset   |  SRE Length   |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |                        Routing Information ...
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *
     * https://datatracker.ietf.org/doc/html/rfc1701
     * https://datatracker.ietf.org/doc/html/rfc2890
     */

    /*
     * Enhanced GRE header (Version 1)
     *
     *   0                   1                   2                   3
     *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |C|R|K|S|s|Recur|A| Flags | Ver |         Protocol Type         |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |    Key (HW) Payload Length    |       Key (LW) Call ID        |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |                  Sequence Number (Optional)                   |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *  |               Acknowledgment Number (Optional)                |
     *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     *
     * https://datatracker.ietf.org/doc/html/rfc2637
     */

/* bit positions for flags in header */
#define GRE_CHECKSUM 0x8000
#define GRE_ROUTING 0x4000
#define GRE_KEY 0x2000
#define GRE_SEQUENCE 0x1000
#define GRE_STRICTSOURCE 0x0800
#define GRE_RECURSION 0x0700
#define GRE_ACK 0x0080          /* only in special PPTPized GRE header */
#define GRE_RESERVED_PPP 0x0078 /* only in special PPTPized GRE header */
#define GRE_RESERVED 0x00F8
#define GRE_VERSION 0x0007

    if (len < 4)
    {
        return 0;
    }

    struct SRE
    {
        uint16_t address_family;
        uint8_t sre_offset;
        uint8_t sre_length;
    } __attribute__((__packed__));

    uint16_t sre_size = sizeof(struct SRE);
    const uint16_t *gre = (const uint16_t *)data;
    uint16_t flags = ntohs(gre[0]);
    uint16_t version = flags & GRE_VERSION;
    uint16_t hdr_offset = 0;

    if (version == 0)
    {
        hdr_offset = 4;

        if ((flags & GRE_CHECKSUM) || (flags & GRE_ROUTING))
        {
            hdr_offset += 4;
        }
        if (flags & GRE_KEY)
        {
            hdr_offset += 4;
        }
        if (flags & GRE_SEQUENCE)
        {
            hdr_offset += 4;
        }
        if (flags & GRE_ROUTING)
        {
            while (hdr_offset + sre_size <= len)
            {
                struct SRE *sre = (struct SRE *)((char *)data + hdr_offset);
                if (sre->sre_length == 0)
                {
                    hdr_offset += sre_size;
                    break;
                }
                else
                {
                    hdr_offset += sre_size + sre->sre_length;
                }
            }
        }
    }

    if (version == 1)
    {
        hdr_offset = 8;
        if (flags & GRE_SEQUENCE)
        {
            hdr_offset += 4;
        }
        if (flags & GRE_ACK)
        {
            hdr_offset += 4;
        }
    }

    if (hdr_offset > len)
    {
        return 0;
    }

    return hdr_offset;
}

static inline const char *parse_ether(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < sizeof(struct ethhdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_ETHER);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t next_proto = ntohs(((struct ethhdr *)data)->h_proto);
    SET_LAYER(handler, layer, LAYER_TYPE_ETHER, sizeof(struct ethhdr), data, len);

    return parse_l3(handler, next_proto, layer->pld_ptr, layer->pld_len);
}

static inline const char *parse_ppp(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < 4))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_PPP);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t next_proto = ntohs(*((uint16_t *)data + 1));
    SET_LAYER(handler, layer, LAYER_TYPE_PPP, 4, data, len);

    switch (next_proto)
    {
    case PPP_IP:
        return parse_ipv4(handler, layer->pld_ptr, layer->pld_len);
    case PPP_IPV6:
        return parse_ipv6(handler, layer->pld_ptr, layer->pld_len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("ppp", next_proto);
        return layer->pld_ptr;
    }
}

static inline const char *parse_vlan(struct packet *handler, const char *data, uint16_t len)
{
    struct vlan_hdr
    {
        uint16_t vlan_cfi;
        uint16_t protocol;
    } __attribute__((__packed__));

    if (unlikely(len < sizeof(struct vlan_hdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_VLAN);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t next_proto = ntohs(((struct vlan_hdr *)data)->protocol);
    SET_LAYER(handler, layer, LAYER_TYPE_VLAN, sizeof(struct vlan_hdr), data, len);

    return parse_l3(handler, next_proto, layer->pld_ptr, layer->pld_len);
}

static inline const char *parse_pppoe_ses(struct packet *handler, const char *data, uint16_t len)
{
#define PPPOE_TYPE_IPV4 0x2100
#define PPPOE_TYPE_IPV6 0x5700

    if (unlikely(len < 8))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_PPPOE);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t next_proto = *((uint16_t *)data + 3);
    SET_LAYER(handler, layer, LAYER_TYPE_PPPOE, 8, data, len);

    switch (next_proto)
    {
    case PPPOE_TYPE_IPV4:
        return parse_ipv4(handler, layer->pld_ptr, layer->pld_len);
    case PPPOE_TYPE_IPV6:
        return parse_ipv6(handler, layer->pld_ptr, layer->pld_len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("pppoe", next_proto);
        return layer->pld_ptr;
    }
}

static inline const char *parse_mpls(struct packet *handler, const char *data, uint16_t len)
{
    /*
     * MPLS Format
     * 0                   1                   2                   3
     * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     * |                Label                  | Exp |S|       TTL     |
     * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     * Label : Label Value      20 bits
     * Exp   : Experimental Use  3 bits
     * S     : Bottom of Stack   1 bit
     * TTL   : Time to Live      8 bits
     */

#define MPLS_LABEL_MASK (0xFFFFF000)
#define MPLS_EXP_MASK (0x00000E00)
#define MPLS_BLS_MASK (0x00000100)
#define MPLS_TTL_MASK (0x000000FF)

    enum mpls_next_proto
    {
        MPLS_NEXT_PROTO_ETHER = 0x0,
        MPLS_NEXT_PROTO_MPLS = 0x1,
        MPLS_NEXT_PROTO_IPV4 = 0x4,
        MPLS_NEXT_PROTO_IPV6 = 0x6,
    };

    // 4 + 1
    if (unlikely(len < 5))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_MPLS);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }

    uint16_t hdr_len = 4;
    uint32_t *hdr = (uint32_t *)data;
    unsigned int mpls_bls = (ntohl(*hdr) & MPLS_BLS_MASK) >> 8;
    enum mpls_next_proto next_proto;
    if (mpls_bls == 1)
    {
        switch ((((uint8_t *)(data + 4))[0]) >> 4)
        {
        case 0:
            /*
             * PW Ethernet Control Word
             * 0                   1                   2                   3
             * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
             * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
             * |0 0 0 0|   Reserved            |       Sequence Number         |
             * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
             * Reference: https://tools.ietf.org/html/rfc4448
             */
            hdr_len += 4; // skip PW Ethernet Control Word
            next_proto = MPLS_NEXT_PROTO_ETHER;
            break;
        case 4:
            next_proto = MPLS_NEXT_PROTO_IPV4;
            break;
        case 6:
            next_proto = MPLS_NEXT_PROTO_IPV6;
            break;
        default:
            next_proto = MPLS_NEXT_PROTO_ETHER;
            break;
        }
    }
    else
    {
        next_proto = MPLS_NEXT_PROTO_MPLS;
    }
    SET_LAYER(handler, layer, LAYER_TYPE_MPLS, hdr_len, data, len);

    switch (next_proto)
    {
    case MPLS_NEXT_PROTO_IPV4:
        return parse_ipv4(handler, layer->pld_ptr, layer->pld_len);
    case MPLS_NEXT_PROTO_IPV6:
        return parse_ipv6(handler, layer->pld_ptr, layer->pld_len);
    case MPLS_NEXT_PROTO_ETHER:
        return parse_ether(handler, layer->pld_ptr, layer->pld_len);
    case MPLS_NEXT_PROTO_MPLS:
        return parse_mpls(handler, layer->pld_ptr, layer->pld_len);
    default:
        // unreachable
        return layer->pld_ptr;
    }
}

static inline const char *parse_ipv4(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < sizeof(struct ip)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_IPV4);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    struct ip *hdr = (struct ip *)data;
    uint8_t next_proto = hdr->ip_p;
    uint16_t hdr_len = (hdr->ip_hl & 0xf) * 4u;
    SET_LAYER(handler, layer, LAYER_TYPE_IPV4, hdr_len, data, len);

    return parse_l4(handler, next_proto, layer->pld_ptr, layer->pld_len);
}

static inline const char *parse_ipv6(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < sizeof(struct ip6_hdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_IPV6);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint8_t next_proto = ((struct ip6_hdr *)data)->ip6_nxt;
    SET_LAYER(handler, layer, LAYER_TYPE_IPV6, sizeof(struct ip6_hdr), data, len);

    return parse_l4(handler, next_proto, layer->pld_ptr, layer->pld_len);
}

static inline const char *parse_gre(struct packet *handler, const char *data, uint16_t len)
{
#define GRE_PRO_IPV4 (0x0800)
#define GRE_PRO_IPV6 (0x86DD)
#define GRE_PRO_ARP (0x0806)
#define GRE_PRO_PPP (0x880B)

    uint16_t hdr_len = gre_header_get_length(data, len);
    if (unlikely(hdr_len == 0))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_GRE);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t next_proto = ntohs(*((uint16_t *)data + 1));
    SET_LAYER(handler, layer, LAYER_TYPE_GRE, hdr_len, data, len);

    switch (next_proto)
    {
    case GRE_PRO_IPV4:
        return parse_ipv4(handler, layer->pld_ptr, layer->pld_len);
    case GRE_PRO_IPV6:
        return parse_ipv6(handler, layer->pld_ptr, layer->pld_len);
    case GRE_PRO_PPP:
        return parse_ppp(handler, layer->pld_ptr, layer->pld_len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("gre", next_proto);
        return layer->pld_ptr;
    }
}

static inline const char *parse_udp(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < sizeof(struct udphdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_UDP);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    struct udphdr *hdr = (struct udphdr *)data;
    SET_LAYER(handler, layer, LAYER_TYPE_UDP, sizeof(struct udphdr), data, len);

    // VXLAN_DPORT 4789
    if (ntohs(hdr->uh_dport) == 4789)
    {
        return parse_vxlan(handler, layer->pld_ptr, layer->pld_len);
    }

    // GTP1U_PORT 2152
    if (ntohs(hdr->uh_dport) == 2152 || ntohs(hdr->uh_sport) == 2152)
    {
        return parse_gtpv1_u(handler, layer->pld_ptr, layer->pld_len);
    }

    return layer->pld_ptr;
}

static inline const char *parse_tcp(struct packet *handler, const char *data, uint16_t len)
{
    if (unlikely(len < sizeof(struct tcphdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_TCP);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint16_t hdr_len = ((struct tcphdr *)data)->th_off << 2;
    SET_LAYER(handler, layer, LAYER_TYPE_TCP, hdr_len, data, len);

    return layer->pld_ptr;
}

static inline const char *parse_vxlan(struct packet *handler, const char *data, uint16_t len)
{
    struct vxlan_hdr
    {
        uint8_t flags[2];
        uint16_t gdp;
        uint8_t vni[3];
        uint8_t reserved;
    } __attribute__((__packed__));

    if (unlikely(len < sizeof(struct vxlan_hdr)))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_VXLAN);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    SET_LAYER(handler, layer, LAYER_TYPE_VXLAN, sizeof(struct vxlan_hdr), data, len);

    return parse_ether(handler, layer->pld_ptr, layer->pld_len);
}

static inline const char *parse_gtpv1_u(struct packet *handler, const char *data, uint16_t len)
{
    uint16_t hdr_len = gtp_header_get_length(data, len);
    if (unlikely(hdr_len == 0))
    {
        PACKET_LOG_DATA_INSUFFICIENCY(LAYER_TYPE_GTPV1_U);
        return data;
    }

    struct layer_record *layer = get_layer(handler);
    if (unlikely(layer == NULL))
    {
        return data;
    }
    uint8_t next_proto = (((const uint8_t *)(data + hdr_len))[0]) >> 4;
    SET_LAYER(handler, layer, LAYER_TYPE_GTPV1_U, hdr_len, data, len);

    switch (next_proto)
    {
    case 4:
        return parse_ipv4(handler, layer->pld_ptr, layer->pld_len);
    case 6:
        return parse_ipv6(handler, layer->pld_ptr, layer->pld_len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("gtp", next_proto);
        return layer->pld_ptr;
    }
}

static inline const char *parse_l3(struct packet *handler, uint16_t next_proto, const char *data, uint16_t len)
{
    switch (next_proto)
    {
    case ETH_P_8021Q:
    case ETH_P_8021AD:
        return parse_vlan(handler, data, len);
    case ETH_P_IP:
        return parse_ipv4(handler, data, len);
    case ETH_P_IPV6:
        return parse_ipv6(handler, data, len);
    case ETH_P_PPP_SES:
        return parse_pppoe_ses(handler, data, len);
    case ETH_P_MPLS_UC:
        return parse_mpls(handler, data, len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("l3", next_proto);
        return data;
    }
}

static inline const char *parse_l4(struct packet *handler, uint8_t next_proto, const char *data, uint16_t len)
{
    switch (next_proto)
    {
    case IPPROTO_TCP:
        return parse_tcp(handler, data, len);
    case IPPROTO_UDP:
        return parse_udp(handler, data, len);
    case IPPROTO_IPIP:
        return parse_ipv4(handler, data, len);
    case IPPROTO_IPV6:
        return parse_ipv6(handler, data, len);
    case IPPROTO_GRE:
        return parse_gre(handler, data, len);
    default:
        PACKET_LOG_UNSUPPORT_PROTO("l4", next_proto);
        return data;
    }
}

/******************************************************************************
 * Public API
 ******************************************************************************/

// return innermost payload
const char *packet_parse(struct packet *handler, const char *data, uint16_t len)
{
    handler->layers_used = 0;
    handler->layers_size = PACKET_MAX_LAYERS;
    handler->data_ptr = data;
    handler->data_len = len;

    return parse_ether(handler, data, len);
}

void packet_print(const struct packet *handler)
{
    if (handler == NULL)
    {
        return;
    }

    printf("packet: %p, data_ptr: %p, data_len: %u, layers_used: %u, layers_size: %u\n",
           handler, handler->data_ptr, handler->data_len,
           handler->layers_used, handler->layers_size);
    for (uint8_t i = 0; i < handler->layers_used; i++)
    {
        const struct layer_record *layer = &handler->layers[i];
        printf("  layer[%u]: %p, type: %s, hdr_offset: %u, hdr_ptr: %p, hdr_len: %u, pld_ptr: %p, pld_len: %u\n",
               i, layer, layer_type_tostring(layer->type), layer->hdr_offset,
               layer->hdr_ptr, layer->hdr_len, layer->pld_ptr, layer->pld_len);
    }
}

// return  0 : found
// return -1 : not found
int packet_get_innermost_four_tuple(const struct packet *handler, struct four_tuple *tuple)
{
    const struct layer_record *layer_l3 = NULL;
    const struct layer_record *layer_l4 = NULL;
    const struct layer_record *layer = NULL;

    for (int8_t i = handler->layers_used - 1; i >= 0; i--)
    {
        layer = &handler->layers[i];

        // first get L4 layer
        if (layer->type & LAYER_TYPE_L4)
        {
            layer_l4 = layer;
            continue;
        }

        // second get L3 layer
        if (layer->type & LAYER_TYPE_L3)
        {
            layer_l3 = layer;
            break;
        }
    }

    if (layer_l3 && layer_l4)
    {
        set_four_tuple((const char *)handler->data_ptr + layer_l3->hdr_offset, layer_l3->type, tuple);
        set_four_tuple((const char *)handler->data_ptr + layer_l4->hdr_offset, layer_l4->type, tuple);
        return 0;
    }
    else
    {
        return -1;
    }
}

// return  0 : found
// return -1 : not found
int packet_get_outermost_four_tuple(const struct packet *handler, struct four_tuple *tuple)
{
    const struct layer_record *layer_l3 = NULL;
    const struct layer_record *layer_l4 = NULL;
    const struct layer_record *layer = NULL;

    for (int8_t i = 0; i < handler->layers_used; i++)
    {
        layer = &handler->layers[i];

        // first get L3 layer
        if (layer->type & LAYER_TYPE_L3)
        {
            layer_l3 = layer;
            continue;
        }

        // second get L4 layer
        if (layer->type & LAYER_TYPE_L4)
        {
            layer_l4 = layer;
            break;
        }
    }

    if (layer_l3 && layer_l4)
    {
        set_four_tuple((const char *)handler->data_ptr + layer_l3->hdr_offset, layer_l3->type, tuple);
        set_four_tuple((const char *)handler->data_ptr + layer_l4->hdr_offset, layer_l4->type, tuple);
        return 0;
    }
    else
    {
        return -1;
    }
}

// return  0 : found
// return -1 : not found
int packet_get_innermost_two_tuple(const struct packet *handler, struct two_tuple *tuple)
{
    const struct layer_record *layer = NULL;

    for (int8_t i = handler->layers_used - 1; i >= 0; i--)
    {
        layer = &handler->layers[i];

        if (layer->type & LAYER_TYPE_L3)
        {
            set_two_tuple((const char *)handler->data_ptr + layer->hdr_offset, layer->type, tuple);
            return 0;
        }
    }

    return -1;
}

// return  0 : found
// return -1 : not found
int packet_get_outermost_two_tuple(const struct packet *handler, struct two_tuple *tuple)
{
    const struct layer_record *layer = NULL;

    for (int8_t i = 0; i < handler->layers_used; i++)
    {
        layer = &handler->layers[i];

        if (layer->type & LAYER_TYPE_L3)
        {
            set_two_tuple((const char *)handler->data_ptr + layer->hdr_offset, layer->type, tuple);
            return 0;
        }
    }

    return -1;
}

const struct layer_record *packet_get_innermost_layer(const struct packet *handler, enum layer_type type)
{
    const struct layer_record *layer = NULL;

    for (int8_t i = handler->layers_used - 1; i >= 0; i--)
    {
        layer = &handler->layers[i];
        if (layer->type & type)
        {
            return layer;
        }
    }

    return NULL;
}

const struct layer_record *packet_get_outermost_layer(const struct packet *handler, enum layer_type type)
{
    const struct layer_record *layer = NULL;

    for (int8_t i = 0; i < handler->layers_used; i++)
    {
        layer = &handler->layers[i];
        if (layer->type & type)
        {
            return layer;
        }
    }

    return NULL;
}

// direction 1: E2I
// direction 0: I2E
uint64_t packet_get_hash(const struct packet *handler, enum ldbc_method method, int direction)
{
    uint64_t temp = 0;
    uint64_t hash_value = 1;

    int inner_addr_len = 0;
    int outer_addr_len = 0;
    const char *inner_src_addr = NULL;
    const char *inner_dst_addr = NULL;
    const char *outer_src_addr = NULL;
    const char *outer_dst_addr = NULL;

    struct two_tuple inner_addr;
    struct two_tuple outer_addr;

    if (handler == NULL)
    {
        return hash_value;
    }

    if (packet_get_innermost_two_tuple(handler, &inner_addr) == -1)
    {
        return hash_value;
    }

    if (packet_get_outermost_two_tuple(handler, &outer_addr) == -1)
    {
        return hash_value;
    }

    inner_src_addr = (const char *)two_tuple_get_src_address(&inner_addr);
    inner_dst_addr = (const char *)two_tuple_get_dst_address(&inner_addr);
    inner_addr_len = two_tuple_get_address_length(&inner_addr);

    outer_src_addr = (const char *)two_tuple_get_src_address(&outer_addr);
    outer_dst_addr = (const char *)two_tuple_get_dst_address(&outer_addr);
    outer_addr_len = two_tuple_get_address_length(&outer_addr);

    switch (method)
    {
    case LDBC_METHOD_HASH_INT_IP:
        if (direction)
        {
            // direction 1: E2I
            HASH_VALUE(outer_dst_addr, outer_addr_len, hash_value);
        }
        else
        {
            // direction 0: I2E
            HASH_VALUE(outer_src_addr, outer_addr_len, hash_value);
        }
        break;
    case LDBC_METHOD_HASH_EXT_IP:
        if (direction)
        {
            // direction 1: E2I
            HASH_VALUE(outer_src_addr, outer_addr_len, hash_value);
        }
        else
        {
            // direction 0: I2E
            HASH_VALUE(outer_dst_addr, outer_addr_len, hash_value);
        }
        break;
    case LDBC_METHOD_HASH_INT_IP_AND_EXT_IP:
        HASH_VALUE(outer_src_addr, outer_addr_len, hash_value);
        HASH_VALUE(outer_dst_addr, outer_addr_len, temp);
        hash_value = hash_value ^ temp;
        break;
    case LDBC_METHOD_HASH_INNERMOST_INT_IP:
        if (direction)
        {
            // direction 1: E2I
            HASH_VALUE(inner_dst_addr, inner_addr_len, hash_value);
        }
        else
        {
            // direction 0: I2E
            HASH_VALUE(inner_src_addr, inner_addr_len, hash_value);
        }
        break;
    case LDBC_METHOD_HASH_INNERMOST_EXT_IP:
        if (direction)
        {
            // direction 1: E2I
            HASH_VALUE(inner_src_addr, inner_addr_len, hash_value);
        }
        else
        {
            // direction 0: I2E
            HASH_VALUE(inner_dst_addr, inner_addr_len, hash_value);
        }
        break;
    default:
        return hash_value;
    }

#if 0
    char *inner_addr_str = two_tuple_tostring(&inner_addr);
    char *outer_addr_str = two_tuple_tostring(&outer_addr);
    printf("%s: outer_addr: %s, inner_addr: %s, dir: %s, hash_method: %s, hash_value: %lu\n",
           LOG_PACKET, outer_addr_str, inner_addr_str, (direction ? "E2I" : "I2E"), ldbc_method_tostring(method), hash_value);
    free(inner_addr_str);
    free(outer_addr_str);
#endif

    return hash_value;
}