tools/testing/selftests/bpf/test_xdp.c

   1 /* Copyright (c) 2016,2017 Facebook
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  */
   7 #include <stddef.h>
   8 #include <string.h>
   9 #include <linux/bpf.h>
  10 #include <linux/if_ether.h>
  11 #include <linux/if_packet.h>
  12 #include <linux/ip.h>
  13 #include <linux/ipv6.h>
  14 #include <linux/in.h>
  15 #include <linux/udp.h>
  16 #include <linux/tcp.h>
  17 #include <linux/pkt_cls.h>
  18 #include <sys/socket.h>
  19 #include "bpf_helpers.h"
  20 #include "bpf_endian.h"
  21 #include "test_iptunnel_common.h"
  22
  23 int _version SEC("version") = 1;
  24
  25 struct bpf_map_def SEC("maps") rxcnt = {
  26         .type = BPF_MAP_TYPE_PERCPU_ARRAY,
  27         .key_size = sizeof(__u32),
  28         .value_size = sizeof(__u64),
  29         .max_entries = 256,
  30 };
  31
  32 struct bpf_map_def SEC("maps") vip2tnl = {
  33         .type = BPF_MAP_TYPE_HASH,
  34         .key_size = sizeof(struct vip),
  35         .value_size = sizeof(struct iptnl_info),
  36         .max_entries = MAX_IPTNL_ENTRIES,
  37 };
  38
  39 static __always_inline void count_tx(__u32 protocol)
  40 {
  41         __u64 *rxcnt_count;
  42
  43         rxcnt_count = bpf_map_lookup_elem(&rxcnt, &protocol);
  44         if (rxcnt_count)
  45                 *rxcnt_count += 1;
  46 }
  47
  48 static __always_inline int get_dport(void *trans_data, void *data_end,
  49                                      __u8 protocol)
  50 {
  51         struct tcphdr *th;
  52         struct udphdr *uh;
  53
  54         switch (protocol) {
  55         case IPPROTO_TCP:
  56                 th = (struct tcphdr *)trans_data;
  57                 if (th + 1 > data_end)
  58                         return -1;
  59                 return th->dest;
  60         case IPPROTO_UDP:
  61                 uh = (struct udphdr *)trans_data;
  62                 if (uh + 1 > data_end)
  63                         return -1;
  64                 return uh->dest;
  65         default:
  66                 return 0;
  67         }
  68 }
  69
  70 static __always_inline void set_ethhdr(struct ethhdr *new_eth,
  71                                        const struct ethhdr *old_eth,
  72                                        const struct iptnl_info *tnl,
  73                                        __be16 h_proto)
  74 {
  75         memcpy(new_eth->h_source, old_eth->h_dest, sizeof(new_eth->h_source));
  76         memcpy(new_eth->h_dest, tnl->dmac, sizeof(new_eth->h_dest));
  77         new_eth->h_proto = h_proto;
  78 }
  79
  80 static __always_inline int handle_ipv4(struct xdp_md *xdp)
  81 {
  82         void *data_end = (void *)(long)xdp->data_end;
  83         void *data = (void *)(long)xdp->data;
  84         struct iptnl_info *tnl;
  85         struct ethhdr *new_eth;
  86         struct ethhdr *old_eth;
  87         struct iphdr *iph = data + sizeof(struct ethhdr);
  88         __u16 *next_iph;
  89         __u16 payload_len;
  90         struct vip vip = {};
  91         int dport;
  92         __u32 csum = 0;
  93         int i;
  94
  95         if (iph + 1 > data_end)
  96                 return XDP_DROP;
  97
  98         dport = get_dport(iph + 1, data_end, iph->protocol);
  99         if (dport == -1)
 100                 return XDP_DROP;
 101
 102         vip.protocol = iph->protocol;
 103         vip.family = AF_INET;
 104         vip.daddr.v4 = iph->daddr;
 105         vip.dport = dport;
 106         payload_len = bpf_ntohs(iph->tot_len);
 107
 108         tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
 109         /* It only does v4-in-v4 */
 110         if (!tnl || tnl->family != AF_INET)
 111                 return XDP_PASS;
 112
 113         if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr)))
 114                 return XDP_DROP;
 115
 116         data = (void *)(long)xdp->data;
 117         data_end = (void *)(long)xdp->data_end;
 118
 119         new_eth = data;
 120         iph = data + sizeof(*new_eth);
 121         old_eth = data + sizeof(*iph);
 122
 123         if (new_eth + 1 > data_end ||
 124             old_eth + 1 > data_end ||
 125             iph + 1 > data_end)
 126                 return XDP_DROP;
 127
 128         set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IP));
 129
 130         iph->version = 4;
 131         iph->ihl = sizeof(*iph) >> 2;
 132         iph->frag_off = 0;
 133         iph->protocol = IPPROTO_IPIP;
 134         iph->check = 0;
 135         iph->tos = 0;
 136         iph->tot_len = bpf_htons(payload_len + sizeof(*iph));
 137         iph->daddr = tnl->daddr.v4;
 138         iph->saddr = tnl->saddr.v4;
 139         iph->ttl = 8;
 140
 141         next_iph = (__u16 *)iph;
 142 #pragma clang loop unroll(full)
 143         for (i = 0; i < sizeof(*iph) >> 1; i++)
 144                 csum += *next_iph++;
 145
 146         iph->check = ~((csum & 0xffff) + (csum >> 16));
 147
 148         count_tx(vip.protocol);
 149
 150         return XDP_TX;
 151 }
 152
 153 static __always_inline int handle_ipv6(struct xdp_md *xdp)
 154 {
 155         void *data_end = (void *)(long)xdp->data_end;
 156         void *data = (void *)(long)xdp->data;
 157         struct iptnl_info *tnl;
 158         struct ethhdr *new_eth;
 159         struct ethhdr *old_eth;
 160         struct ipv6hdr *ip6h = data + sizeof(struct ethhdr);
 161         __u16 payload_len;
 162         struct vip vip = {};
 163         int dport;
 164
 165         if (ip6h + 1 > data_end)
 166                 return XDP_DROP;
 167
 168         dport = get_dport(ip6h + 1, data_end, ip6h->nexthdr);
 169         if (dport == -1)
 170                 return XDP_DROP;
 171
 172         vip.protocol = ip6h->nexthdr;
 173         vip.family = AF_INET6;
 174         memcpy(vip.daddr.v6, ip6h->daddr.s6_addr32, sizeof(vip.daddr));
 175         vip.dport = dport;
 176         payload_len = ip6h->payload_len;
 177
 178         tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
 179         /* It only does v6-in-v6 */
 180         if (!tnl || tnl->family != AF_INET6)
 181                 return XDP_PASS;
 182
 183         if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr)))
 184                 return XDP_DROP;
 185
 186         data = (void *)(long)xdp->data;
 187         data_end = (void *)(long)xdp->data_end;
 188
 189         new_eth = data;
 190         ip6h = data + sizeof(*new_eth);
 191         old_eth = data + sizeof(*ip6h);
 192
 193         if (new_eth + 1 > data_end || old_eth + 1 > data_end ||
 194             ip6h + 1 > data_end)
 195                 return XDP_DROP;
 196
 197         set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IPV6));
 198
 199         ip6h->version = 6;
 200         ip6h->priority = 0;
 201         memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
 202         ip6h->payload_len = bpf_htons(bpf_ntohs(payload_len) + sizeof(*ip6h));
 203         ip6h->nexthdr = IPPROTO_IPV6;
 204         ip6h->hop_limit = 8;
 205         memcpy(ip6h->saddr.s6_addr32, tnl->saddr.v6, sizeof(tnl->saddr.v6));
 206         memcpy(ip6h->daddr.s6_addr32, tnl->daddr.v6, sizeof(tnl->daddr.v6));
 207
 208         count_tx(vip.protocol);
 209
 210         return XDP_TX;
 211 }
 212
 213 SEC("xdp_tx_iptunnel")
 214 int _xdp_tx_iptunnel(struct xdp_md *xdp)
 215 {
 216         void *data_end = (void *)(long)xdp->data_end;
 217         void *data = (void *)(long)xdp->data;
 218         struct ethhdr *eth = data;
 219         __u16 h_proto;
 220
 221         if (eth + 1 > data_end)
 222                 return XDP_DROP;
 223
 224         h_proto = eth->h_proto;
 225
 226         if (h_proto == bpf_htons(ETH_P_IP))
 227                 return handle_ipv4(xdp);
 228         else if (h_proto == bpf_htons(ETH_P_IPV6))
 229
 230                 return handle_ipv6(xdp);
 231         else
 232                 return XDP_DROP;
 233 }
 234
 235 char _license[] SEC("license") = "GPL";