bpf/progs/test_xdp_noinline.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 // Copyright (c) 2017 Facebook
0003 #include <stddef.h>
0004 #include <stdbool.h>
0005 #include <string.h>
0006 #include <linux/pkt_cls.h>
0007 #include <linux/bpf.h>
0008 #include <linux/in.h>
0009 #include <linux/if_ether.h>
0010 #include <linux/ip.h>
0011 #include <linux/ipv6.h>
0012 #include <linux/icmp.h>
0013 #include <linux/icmpv6.h>
0014 #include <linux/tcp.h>
0015 #include <linux/udp.h>
0016 #include <bpf/bpf_helpers.h>
0017 #include <bpf/bpf_endian.h>
0018
0019 static __always_inline __u32 rol32(__u32 word, unsigned int shift)
0020 {
0021     return (word << shift) | (word >> ((-shift) & 31));
0022 }
0023
0024 /* copy paste of jhash from kernel sources to make sure llvm
0025  * can compile it into valid sequence of bpf instructions
0026  */
0027 #define __jhash_mix(a, b, c)            \
0028 {                       \
0029     a -= c;  a ^= rol32(c, 4);  c += b; \
0030     b -= a;  b ^= rol32(a, 6);  a += c; \
0031     c -= b;  c ^= rol32(b, 8);  b += a; \
0032     a -= c;  a ^= rol32(c, 16); c += b; \
0033     b -= a;  b ^= rol32(a, 19); a += c; \
0034     c -= b;  c ^= rol32(b, 4);  b += a; \
0035 }
0036
0037 #define __jhash_final(a, b, c)          \
0038 {                       \
0039     c ^= b; c -= rol32(b, 14);      \
0040     a ^= c; a -= rol32(c, 11);      \
0041     b ^= a; b -= rol32(a, 25);      \
0042     c ^= b; c -= rol32(b, 16);      \
0043     a ^= c; a -= rol32(c, 4);       \
0044     b ^= a; b -= rol32(a, 14);      \
0045     c ^= b; c -= rol32(b, 24);      \
0046 }
0047
0048 #define JHASH_INITVAL       0xdeadbeef
0049
0050 typedef unsigned int u32;
0051
0052 static __noinline
0053 u32 jhash(const void *key, u32 length, u32 initval)
0054 {
0055     u32 a, b, c;
0056     const unsigned char *k = key;
0057
0058     a = b = c = JHASH_INITVAL + length + initval;
0059
0060     while (length > 12) {
0061         a += *(u32 *)(k);
0062         b += *(u32 *)(k + 4);
0063         c += *(u32 *)(k + 8);
0064         __jhash_mix(a, b, c);
0065         length -= 12;
0066         k += 12;
0067     }
0068     switch (length) {
0069     case 12: c += (u32)k[11]<<24;
0070     case 11: c += (u32)k[10]<<16;
0071     case 10: c += (u32)k[9]<<8;
0072     case 9:  c += k[8];
0073     case 8:  b += (u32)k[7]<<24;
0074     case 7:  b += (u32)k[6]<<16;
0075     case 6:  b += (u32)k[5]<<8;
0076     case 5:  b += k[4];
0077     case 4:  a += (u32)k[3]<<24;
0078     case 3:  a += (u32)k[2]<<16;
0079     case 2:  a += (u32)k[1]<<8;
0080     case 1:  a += k[0];
0081          __jhash_final(a, b, c);
0082     case 0: /* Nothing left to add */
0083         break;
0084     }
0085
0086     return c;
0087 }
0088
0089 __noinline
0090 u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
0091 {
0092     a += initval;
0093     b += initval;
0094     c += initval;
0095     __jhash_final(a, b, c);
0096     return c;
0097 }
0098
0099 __noinline
0100 u32 jhash_2words(u32 a, u32 b, u32 initval)
0101 {
0102     return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
0103 }
0104
0105 struct flow_key {
0106     union {
0107         __be32 src;
0108         __be32 srcv6[4];
0109     };
0110     union {
0111         __be32 dst;
0112         __be32 dstv6[4];
0113     };
0114     union {
0115         __u32 ports;
0116         __u16 port16[2];
0117     };
0118     __u8 proto;
0119 };
0120
0121 struct packet_description {
0122     struct flow_key flow;
0123     __u8 flags;
0124 };
0125
0126 struct ctl_value {
0127     union {
0128         __u64 value;
0129         __u32 ifindex;
0130         __u8 mac[6];
0131     };
0132 };
0133
0134 struct vip_definition {
0135     union {
0136         __be32 vip;
0137         __be32 vipv6[4];
0138     };
0139     __u16 port;
0140     __u16 family;
0141     __u8 proto;
0142 };
0143
0144 struct vip_meta {
0145     __u32 flags;
0146     __u32 vip_num;
0147 };
0148
0149 struct real_pos_lru {
0150     __u32 pos;
0151     __u64 atime;
0152 };
0153
0154 struct real_definition {
0155     union {
0156         __be32 dst;
0157         __be32 dstv6[4];
0158     };
0159     __u8 flags;
0160 };
0161
0162 struct lb_stats {
0163     __u64 v2;
0164     __u64 v1;
0165 };
0166
0167 struct {
0168     __uint(type, BPF_MAP_TYPE_HASH);
0169     __uint(max_entries, 512);
0170     __type(key, struct vip_definition);
0171     __type(value, struct vip_meta);
0172 } vip_map SEC(".maps");
0173
0174 struct {
0175     __uint(type, BPF_MAP_TYPE_LRU_HASH);
0176     __uint(max_entries, 300);
0177     __uint(map_flags, 1U << 1);
0178     __type(key, struct flow_key);
0179     __type(value, struct real_pos_lru);
0180 } lru_cache SEC(".maps");
0181
0182 struct {
0183     __uint(type, BPF_MAP_TYPE_ARRAY);
0184     __uint(max_entries, 12 * 655);
0185     __type(key, __u32);
0186     __type(value, __u32);
0187 } ch_rings SEC(".maps");
0188
0189 struct {
0190     __uint(type, BPF_MAP_TYPE_ARRAY);
0191     __uint(max_entries, 40);
0192     __type(key, __u32);
0193     __type(value, struct real_definition);
0194 } reals SEC(".maps");
0195
0196 struct {
0197     __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
0198     __uint(max_entries, 515);
0199     __type(key, __u32);
0200     __type(value, struct lb_stats);
0201 } stats SEC(".maps");
0202
0203 struct {
0204     __uint(type, BPF_MAP_TYPE_ARRAY);
0205     __uint(max_entries, 16);
0206     __type(key, __u32);
0207     __type(value, struct ctl_value);
0208 } ctl_array SEC(".maps");
0209
0210 struct eth_hdr {
0211     unsigned char eth_dest[6];
0212     unsigned char eth_source[6];
0213     unsigned short eth_proto;
0214 };
0215
0216 static __noinline __u64 calc_offset(bool is_ipv6, bool is_icmp)
0217 {
0218     __u64 off = sizeof(struct eth_hdr);
0219     if (is_ipv6) {
0220         off += sizeof(struct ipv6hdr);
0221         if (is_icmp)
0222             off += sizeof(struct icmp6hdr) + sizeof(struct ipv6hdr);
0223     } else {
0224         off += sizeof(struct iphdr);
0225         if (is_icmp)
0226             off += sizeof(struct icmphdr) + sizeof(struct iphdr);
0227     }
0228     return off;
0229 }
0230
0231 static __attribute__ ((noinline))
0232 bool parse_udp(void *data, void *data_end,
0233            bool is_ipv6, struct packet_description *pckt)
0234 {
0235
0236     bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
0237     __u64 off = calc_offset(is_ipv6, is_icmp);
0238     struct udphdr *udp;
0239     udp = data + off;
0240
0241     if (udp + 1 > data_end)
0242         return false;
0243     if (!is_icmp) {
0244         pckt->flow.port16[0] = udp->source;
0245         pckt->flow.port16[1] = udp->dest;
0246     } else {
0247         pckt->flow.port16[0] = udp->dest;
0248         pckt->flow.port16[1] = udp->source;
0249     }
0250     return true;
0251 }
0252
0253 static __attribute__ ((noinline))
0254 bool parse_tcp(void *data, void *data_end,
0255            bool is_ipv6, struct packet_description *pckt)
0256 {
0257
0258     bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
0259     __u64 off = calc_offset(is_ipv6, is_icmp);
0260     struct tcphdr *tcp;
0261
0262     tcp = data + off;
0263     if (tcp + 1 > data_end)
0264         return false;
0265     if (tcp->syn)
0266         pckt->flags |= (1 << 1);
0267     if (!is_icmp) {
0268         pckt->flow.port16[0] = tcp->source;
0269         pckt->flow.port16[1] = tcp->dest;
0270     } else {
0271         pckt->flow.port16[0] = tcp->dest;
0272         pckt->flow.port16[1] = tcp->source;
0273     }
0274     return true;
0275 }
0276
0277 static __attribute__ ((noinline))
0278 bool encap_v6(struct xdp_md *xdp, struct ctl_value *cval,
0279           struct packet_description *pckt,
0280           struct real_definition *dst, __u32 pkt_bytes)
0281 {
0282     struct eth_hdr *new_eth;
0283     struct eth_hdr *old_eth;
0284     struct ipv6hdr *ip6h;
0285     __u32 ip_suffix;
0286     void *data_end;
0287     void *data;
0288
0289     if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr)))
0290         return false;
0291     data = (void *)(long)xdp->data;
0292     data_end = (void *)(long)xdp->data_end;
0293     new_eth = data;
0294     ip6h = data + sizeof(struct eth_hdr);
0295     old_eth = data + sizeof(struct ipv6hdr);
0296     if (new_eth + 1 > data_end ||
0297         old_eth + 1 > data_end || ip6h + 1 > data_end)
0298         return false;
0299     memcpy(new_eth->eth_dest, cval->mac, 6);
0300     memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
0301     new_eth->eth_proto = 56710;
0302     ip6h->version = 6;
0303     ip6h->priority = 0;
0304     memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
0305
0306     ip6h->nexthdr = IPPROTO_IPV6;
0307     ip_suffix = pckt->flow.srcv6[3] ^ pckt->flow.port16[0];
0308     ip6h->payload_len =
0309         bpf_htons(pkt_bytes + sizeof(struct ipv6hdr));
0310     ip6h->hop_limit = 4;
0311
0312     ip6h->saddr.in6_u.u6_addr32[0] = 1;
0313     ip6h->saddr.in6_u.u6_addr32[1] = 2;
0314     ip6h->saddr.in6_u.u6_addr32[2] = 3;
0315     ip6h->saddr.in6_u.u6_addr32[3] = ip_suffix;
0316     memcpy(ip6h->daddr.in6_u.u6_addr32, dst->dstv6, 16);
0317     return true;
0318 }
0319
0320 static __attribute__ ((noinline))
0321 bool encap_v4(struct xdp_md *xdp, struct ctl_value *cval,
0322           struct packet_description *pckt,
0323           struct real_definition *dst, __u32 pkt_bytes)
0324 {
0325
0326     __u32 ip_suffix = bpf_ntohs(pckt->flow.port16[0]);
0327     struct eth_hdr *new_eth;
0328     struct eth_hdr *old_eth;
0329     __u16 *next_iph_u16;
0330     struct iphdr *iph;
0331     __u32 csum = 0;
0332     void *data_end;
0333     void *data;
0334
0335     ip_suffix <<= 15;
0336     ip_suffix ^= pckt->flow.src;
0337     if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr)))
0338         return false;
0339     data = (void *)(long)xdp->data;
0340     data_end = (void *)(long)xdp->data_end;
0341     new_eth = data;
0342     iph = data + sizeof(struct eth_hdr);
0343     old_eth = data + sizeof(struct iphdr);
0344     if (new_eth + 1 > data_end ||
0345         old_eth + 1 > data_end || iph + 1 > data_end)
0346         return false;
0347     memcpy(new_eth->eth_dest, cval->mac, 6);
0348     memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
0349     new_eth->eth_proto = 8;
0350     iph->version = 4;
0351     iph->ihl = 5;
0352     iph->frag_off = 0;
0353     iph->protocol = IPPROTO_IPIP;
0354     iph->check = 0;
0355     iph->tos = 1;
0356     iph->tot_len = bpf_htons(pkt_bytes + sizeof(struct iphdr));
0357     /* don't update iph->daddr, since it will overwrite old eth_proto
0358      * and multiple iterations of bpf_prog_run() will fail
0359      */
0360
0361     iph->saddr = ((0xFFFF0000 & ip_suffix) | 4268) ^ dst->dst;
0362     iph->ttl = 4;
0363
0364     next_iph_u16 = (__u16 *) iph;
0365 #pragma clang loop unroll(full)
0366     for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
0367         csum += *next_iph_u16++;
0368     iph->check = ~((csum & 0xffff) + (csum >> 16));
0369     if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct iphdr)))
0370         return false;
0371     return true;
0372 }
0373
0374 static __attribute__ ((noinline))
0375 bool decap_v6(struct xdp_md *xdp, void **data, void **data_end, bool inner_v4)
0376 {
0377     struct eth_hdr *new_eth;
0378     struct eth_hdr *old_eth;
0379
0380     old_eth = *data;
0381     new_eth = *data + sizeof(struct ipv6hdr);
0382     memcpy(new_eth->eth_source, old_eth->eth_source, 6);
0383     memcpy(new_eth->eth_dest, old_eth->eth_dest, 6);
0384     if (inner_v4)
0385         new_eth->eth_proto = 8;
0386     else
0387         new_eth->eth_proto = 56710;
0388     if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct ipv6hdr)))
0389         return false;
0390     *data = (void *)(long)xdp->data;
0391     *data_end = (void *)(long)xdp->data_end;
0392     return true;
0393 }
0394
0395 static __attribute__ ((noinline))
0396 bool decap_v4(struct xdp_md *xdp, void **data, void **data_end)
0397 {
0398     struct eth_hdr *new_eth;
0399     struct eth_hdr *old_eth;
0400
0401     old_eth = *data;
0402     new_eth = *data + sizeof(struct iphdr);
0403     memcpy(new_eth->eth_source, old_eth->eth_source, 6);
0404     memcpy(new_eth->eth_dest, old_eth->eth_dest, 6);
0405     new_eth->eth_proto = 8;
0406     if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct iphdr)))
0407         return false;
0408     *data = (void *)(long)xdp->data;
0409     *data_end = (void *)(long)xdp->data_end;
0410     return true;
0411 }
0412
0413 static __attribute__ ((noinline))
0414 int swap_mac_and_send(void *data, void *data_end)
0415 {
0416     unsigned char tmp_mac[6];
0417     struct eth_hdr *eth;
0418
0419     eth = data;
0420     memcpy(tmp_mac, eth->eth_source, 6);
0421     memcpy(eth->eth_source, eth->eth_dest, 6);
0422     memcpy(eth->eth_dest, tmp_mac, 6);
0423     return XDP_TX;
0424 }
0425
0426 static __attribute__ ((noinline))
0427 int send_icmp_reply(void *data, void *data_end)
0428 {
0429     struct icmphdr *icmp_hdr;
0430     __u16 *next_iph_u16;
0431     __u32 tmp_addr = 0;
0432     struct iphdr *iph;
0433     __u32 csum1 = 0;
0434     __u32 csum = 0;
0435     __u64 off = 0;
0436
0437     if (data + sizeof(struct eth_hdr)
0438          + sizeof(struct iphdr) + sizeof(struct icmphdr) > data_end)
0439         return XDP_DROP;
0440     off += sizeof(struct eth_hdr);
0441     iph = data + off;
0442     off += sizeof(struct iphdr);
0443     icmp_hdr = data + off;
0444     icmp_hdr->type = 0;
0445     icmp_hdr->checksum += 0x0007;
0446     iph->ttl = 4;
0447     tmp_addr = iph->daddr;
0448     iph->daddr = iph->saddr;
0449     iph->saddr = tmp_addr;
0450     iph->check = 0;
0451     next_iph_u16 = (__u16 *) iph;
0452 #pragma clang loop unroll(full)
0453     for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
0454         csum += *next_iph_u16++;
0455     iph->check = ~((csum & 0xffff) + (csum >> 16));
0456     return swap_mac_and_send(data, data_end);
0457 }
0458
0459 static __attribute__ ((noinline))
0460 int send_icmp6_reply(void *data, void *data_end)
0461 {
0462     struct icmp6hdr *icmp_hdr;
0463     struct ipv6hdr *ip6h;
0464     __be32 tmp_addr[4];
0465     __u64 off = 0;
0466
0467     if (data + sizeof(struct eth_hdr)
0468          + sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr) > data_end)
0469         return XDP_DROP;
0470     off += sizeof(struct eth_hdr);
0471     ip6h = data + off;
0472     off += sizeof(struct ipv6hdr);
0473     icmp_hdr = data + off;
0474     icmp_hdr->icmp6_type = 129;
0475     icmp_hdr->icmp6_cksum -= 0x0001;
0476     ip6h->hop_limit = 4;
0477     memcpy(tmp_addr, ip6h->saddr.in6_u.u6_addr32, 16);
0478     memcpy(ip6h->saddr.in6_u.u6_addr32, ip6h->daddr.in6_u.u6_addr32, 16);
0479     memcpy(ip6h->daddr.in6_u.u6_addr32, tmp_addr, 16);
0480     return swap_mac_and_send(data, data_end);
0481 }
0482
0483 static __attribute__ ((noinline))
0484 int parse_icmpv6(void *data, void *data_end, __u64 off,
0485          struct packet_description *pckt)
0486 {
0487     struct icmp6hdr *icmp_hdr;
0488     struct ipv6hdr *ip6h;
0489
0490     icmp_hdr = data + off;
0491     if (icmp_hdr + 1 > data_end)
0492         return XDP_DROP;
0493     if (icmp_hdr->icmp6_type == 128)
0494         return send_icmp6_reply(data, data_end);
0495     if (icmp_hdr->icmp6_type != 3)
0496         return XDP_PASS;
0497     off += sizeof(struct icmp6hdr);
0498     ip6h = data + off;
0499     if (ip6h + 1 > data_end)
0500         return XDP_DROP;
0501     pckt->flow.proto = ip6h->nexthdr;
0502     pckt->flags |= (1 << 0);
0503     memcpy(pckt->flow.srcv6, ip6h->daddr.in6_u.u6_addr32, 16);
0504     memcpy(pckt->flow.dstv6, ip6h->saddr.in6_u.u6_addr32, 16);
0505     return -1;
0506 }
0507
0508 static __attribute__ ((noinline))
0509 int parse_icmp(void *data, void *data_end, __u64 off,
0510            struct packet_description *pckt)
0511 {
0512     struct icmphdr *icmp_hdr;
0513     struct iphdr *iph;
0514
0515     icmp_hdr = data + off;
0516     if (icmp_hdr + 1 > data_end)
0517         return XDP_DROP;
0518     if (icmp_hdr->type == 8)
0519         return send_icmp_reply(data, data_end);
0520     if ((icmp_hdr->type != 3) || (icmp_hdr->code != 4))
0521         return XDP_PASS;
0522     off += sizeof(struct icmphdr);
0523     iph = data + off;
0524     if (iph + 1 > data_end)
0525         return XDP_DROP;
0526     if (iph->ihl != 5)
0527         return XDP_DROP;
0528     pckt->flow.proto = iph->protocol;
0529     pckt->flags |= (1 << 0);
0530     pckt->flow.src = iph->daddr;
0531     pckt->flow.dst = iph->saddr;
0532     return -1;
0533 }
0534
0535 static __attribute__ ((noinline))
0536 __u32 get_packet_hash(struct packet_description *pckt,
0537               bool hash_16bytes)
0538 {
0539     if (hash_16bytes)
0540         return jhash_2words(jhash(pckt->flow.srcv6, 16, 12),
0541                     pckt->flow.ports, 24);
0542     else
0543         return jhash_2words(pckt->flow.src, pckt->flow.ports,
0544                     24);
0545 }
0546
0547 __attribute__ ((noinline))
0548 static bool get_packet_dst(struct real_definition **real,
0549                struct packet_description *pckt,
0550                struct vip_meta *vip_info,
0551                bool is_ipv6, void *lru_map)
0552 {
0553     struct real_pos_lru new_dst_lru = { };
0554     bool hash_16bytes = is_ipv6;
0555     __u32 *real_pos, hash, key;
0556     __u64 cur_time;
0557
0558     if (vip_info->flags & (1 << 2))
0559         hash_16bytes = 1;
0560     if (vip_info->flags & (1 << 3)) {
0561         pckt->flow.port16[0] = pckt->flow.port16[1];
0562         memset(pckt->flow.srcv6, 0, 16);
0563     }
0564     hash = get_packet_hash(pckt, hash_16bytes);
0565     if (hash != 0x358459b7 /* jhash of ipv4 packet */  &&
0566         hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
0567         return false;
0568     key = 2 * vip_info->vip_num + hash % 2;
0569     real_pos = bpf_map_lookup_elem(&ch_rings, &key);
0570     if (!real_pos)
0571         return false;
0572     key = *real_pos;
0573     *real = bpf_map_lookup_elem(&reals, &key);
0574     if (!(*real))
0575         return false;
0576     if (!(vip_info->flags & (1 << 1))) {
0577         __u32 conn_rate_key = 512 + 2;
0578         struct lb_stats *conn_rate_stats =
0579             bpf_map_lookup_elem(&stats, &conn_rate_key);
0580
0581         if (!conn_rate_stats)
0582             return true;
0583         cur_time = bpf_ktime_get_ns();
0584         if ((cur_time - conn_rate_stats->v2) >> 32 > 0xffFFFF) {
0585             conn_rate_stats->v1 = 1;
0586             conn_rate_stats->v2 = cur_time;
0587         } else {
0588             conn_rate_stats->v1 += 1;
0589             if (conn_rate_stats->v1 >= 1)
0590                 return true;
0591         }
0592         if (pckt->flow.proto == IPPROTO_UDP)
0593             new_dst_lru.atime = cur_time;
0594         new_dst_lru.pos = key;
0595         bpf_map_update_elem(lru_map, &pckt->flow, &new_dst_lru, 0);
0596     }
0597     return true;
0598 }
0599
0600 __attribute__ ((noinline))
0601 static void connection_table_lookup(struct real_definition **real,
0602                     struct packet_description *pckt,
0603                     void *lru_map)
0604 {
0605
0606     struct real_pos_lru *dst_lru;
0607     __u64 cur_time;
0608     __u32 key;
0609
0610     dst_lru = bpf_map_lookup_elem(lru_map, &pckt->flow);
0611     if (!dst_lru)
0612         return;
0613     if (pckt->flow.proto == IPPROTO_UDP) {
0614         cur_time = bpf_ktime_get_ns();
0615         if (cur_time - dst_lru->atime > 300000)
0616             return;
0617         dst_lru->atime = cur_time;
0618     }
0619     key = dst_lru->pos;
0620     *real = bpf_map_lookup_elem(&reals, &key);
0621 }
0622
0623 /* don't believe your eyes!
0624  * below function has 6 arguments whereas bpf and llvm allow maximum of 5
0625  * but since it's _static_ llvm can optimize one argument away
0626  */
0627 __attribute__ ((noinline))
0628 static int process_l3_headers_v6(struct packet_description *pckt,
0629                  __u8 *protocol, __u64 off,
0630                  __u16 *pkt_bytes, void *data,
0631                  void *data_end)
0632 {
0633     struct ipv6hdr *ip6h;
0634     __u64 iph_len;
0635     int action;
0636
0637     ip6h = data + off;
0638     if (ip6h + 1 > data_end)
0639         return XDP_DROP;
0640     iph_len = sizeof(struct ipv6hdr);
0641     *protocol = ip6h->nexthdr;
0642     pckt->flow.proto = *protocol;
0643     *pkt_bytes = bpf_ntohs(ip6h->payload_len);
0644     off += iph_len;
0645     if (*protocol == 45) {
0646         return XDP_DROP;
0647     } else if (*protocol == 59) {
0648         action = parse_icmpv6(data, data_end, off, pckt);
0649         if (action >= 0)
0650             return action;
0651     } else {
0652         memcpy(pckt->flow.srcv6, ip6h->saddr.in6_u.u6_addr32, 16);
0653         memcpy(pckt->flow.dstv6, ip6h->daddr.in6_u.u6_addr32, 16);
0654     }
0655     return -1;
0656 }
0657
0658 __attribute__ ((noinline))
0659 static int process_l3_headers_v4(struct packet_description *pckt,
0660                  __u8 *protocol, __u64 off,
0661                  __u16 *pkt_bytes, void *data,
0662                  void *data_end)
0663 {
0664     struct iphdr *iph;
0665     __u64 iph_len;
0666     int action;
0667
0668     iph = data + off;
0669     if (iph + 1 > data_end)
0670         return XDP_DROP;
0671     if (iph->ihl != 5)
0672         return XDP_DROP;
0673     *protocol = iph->protocol;
0674     pckt->flow.proto = *protocol;
0675     *pkt_bytes = bpf_ntohs(iph->tot_len);
0676     off += 20;
0677     if (iph->frag_off & 65343)
0678         return XDP_DROP;
0679     if (*protocol == IPPROTO_ICMP) {
0680         action = parse_icmp(data, data_end, off, pckt);
0681         if (action >= 0)
0682             return action;
0683     } else {
0684         pckt->flow.src = iph->saddr;
0685         pckt->flow.dst = iph->daddr;
0686     }
0687     return -1;
0688 }
0689
0690 __attribute__ ((noinline))
0691 static int process_packet(void *data, __u64 off, void *data_end,
0692               bool is_ipv6, struct xdp_md *xdp)
0693 {
0694
0695     struct real_definition *dst = NULL;
0696     struct packet_description pckt = { };
0697     struct vip_definition vip = { };
0698     struct lb_stats *data_stats;
0699     struct eth_hdr *eth = data;
0700     void *lru_map = &lru_cache;
0701     struct vip_meta *vip_info;
0702     __u32 lru_stats_key = 513;
0703     __u32 mac_addr_pos = 0;
0704     __u32 stats_key = 512;
0705     struct ctl_value *cval;
0706     __u16 pkt_bytes;
0707     __u64 iph_len;
0708     __u8 protocol;
0709     __u32 vip_num;
0710     int action;
0711
0712     if (is_ipv6)
0713         action = process_l3_headers_v6(&pckt, &protocol, off,
0714                            &pkt_bytes, data, data_end);
0715     else
0716         action = process_l3_headers_v4(&pckt, &protocol, off,
0717                            &pkt_bytes, data, data_end);
0718     if (action >= 0)
0719         return action;
0720     protocol = pckt.flow.proto;
0721     if (protocol == IPPROTO_TCP) {
0722         if (!parse_tcp(data, data_end, is_ipv6, &pckt))
0723             return XDP_DROP;
0724     } else if (protocol == IPPROTO_UDP) {
0725         if (!parse_udp(data, data_end, is_ipv6, &pckt))
0726             return XDP_DROP;
0727     } else {
0728         return XDP_TX;
0729     }
0730
0731     if (is_ipv6)
0732         memcpy(vip.vipv6, pckt.flow.dstv6, 16);
0733     else
0734         vip.vip = pckt.flow.dst;
0735     vip.port = pckt.flow.port16[1];
0736     vip.proto = pckt.flow.proto;
0737     vip_info = bpf_map_lookup_elem(&vip_map, &vip);
0738     if (!vip_info) {
0739         vip.port = 0;
0740         vip_info = bpf_map_lookup_elem(&vip_map, &vip);
0741         if (!vip_info)
0742             return XDP_PASS;
0743         if (!(vip_info->flags & (1 << 4)))
0744             pckt.flow.port16[1] = 0;
0745     }
0746     if (data_end - data > 1400)
0747         return XDP_DROP;
0748     data_stats = bpf_map_lookup_elem(&stats, &stats_key);
0749     if (!data_stats)
0750         return XDP_DROP;
0751     data_stats->v1 += 1;
0752     if (!dst) {
0753         if (vip_info->flags & (1 << 0))
0754             pckt.flow.port16[0] = 0;
0755         if (!(pckt.flags & (1 << 1)) && !(vip_info->flags & (1 << 1)))
0756             connection_table_lookup(&dst, &pckt, lru_map);
0757         if (dst)
0758             goto out;
0759         if (pckt.flow.proto == IPPROTO_TCP) {
0760             struct lb_stats *lru_stats =
0761                 bpf_map_lookup_elem(&stats, &lru_stats_key);
0762
0763             if (!lru_stats)
0764                 return XDP_DROP;
0765             if (pckt.flags & (1 << 1))
0766                 lru_stats->v1 += 1;
0767             else
0768                 lru_stats->v2 += 1;
0769         }
0770         if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6, lru_map))
0771             return XDP_DROP;
0772         data_stats->v2 += 1;
0773     }
0774 out:
0775     cval = bpf_map_lookup_elem(&ctl_array, &mac_addr_pos);
0776     if (!cval)
0777         return XDP_DROP;
0778     if (dst->flags & (1 << 0)) {
0779         if (!encap_v6(xdp, cval, &pckt, dst, pkt_bytes))
0780             return XDP_DROP;
0781     } else {
0782         if (!encap_v4(xdp, cval, &pckt, dst, pkt_bytes))
0783             return XDP_DROP;
0784     }
0785     vip_num = vip_info->vip_num;
0786     data_stats = bpf_map_lookup_elem(&stats, &vip_num);
0787     if (!data_stats)
0788         return XDP_DROP;
0789     data_stats->v1 += 1;
0790     data_stats->v2 += pkt_bytes;
0791
0792     data = (void *)(long)xdp->data;
0793     data_end = (void *)(long)xdp->data_end;
0794     if (data + 4 > data_end)
0795         return XDP_DROP;
0796     *(u32 *)data = dst->dst;
0797     return XDP_DROP;
0798 }
0799
0800 SEC("xdp")
0801 int balancer_ingress_v4(struct xdp_md *ctx)
0802 {
0803     void *data = (void *)(long)ctx->data;
0804     void *data_end = (void *)(long)ctx->data_end;
0805     struct eth_hdr *eth = data;
0806     __u32 eth_proto;
0807     __u32 nh_off;
0808
0809     nh_off = sizeof(struct eth_hdr);
0810     if (data + nh_off > data_end)
0811         return XDP_DROP;
0812     eth_proto = bpf_ntohs(eth->eth_proto);
0813     if (eth_proto == ETH_P_IP)
0814         return process_packet(data, nh_off, data_end, 0, ctx);
0815     else
0816         return XDP_DROP;
0817 }
0818
0819 SEC("xdp")
0820 int balancer_ingress_v6(struct xdp_md *ctx)
0821 {
0822     void *data = (void *)(long)ctx->data;
0823     void *data_end = (void *)(long)ctx->data_end;
0824     struct eth_hdr *eth = data;
0825     __u32 eth_proto;
0826     __u32 nh_off;
0827
0828     nh_off = sizeof(struct eth_hdr);
0829     if (data + nh_off > data_end)
0830         return XDP_DROP;
0831     eth_proto = bpf_ntohs(eth->eth_proto);
0832     if (eth_proto == ETH_P_IPV6)
0833         return process_packet(data, nh_off, data_end, 1, ctx);
0834     else
0835         return XDP_DROP;
0836 }
0837
0838 char _license[] SEC("license") = "GPL";