1 // SPDX-License-Identifier: GPL-2.0 2 // Copyright (c) 2019 Cloudflare Ltd. 3 // Copyright (c) 2020 Isovalent, Inc. 4 5 #include <stddef.h> 6 #include <stdbool.h> 7 #include <string.h> 8 #include <linux/bpf.h> 9 #include <linux/if_ether.h> 10 #include <linux/in.h> 11 #include <linux/ip.h> 12 #include <linux/ipv6.h> 13 #include <linux/pkt_cls.h> 14 #include <linux/tcp.h> 15 #include <sys/socket.h> 16 #include <bpf/bpf_helpers.h> 17 #include <bpf/bpf_endian.h> 18 19 #if defined(IPROUTE2_HAVE_LIBBPF) 20 /* Use a new-style map definition. */ 21 struct { 22 __uint(type, BPF_MAP_TYPE_SOCKMAP); 23 __type(key, int); 24 __type(value, __u64); 25 __uint(pinning, LIBBPF_PIN_BY_NAME); 26 __uint(max_entries, 1); 27 } server_map SEC(".maps"); 28 #else 29 /* Pin map under /sys/fs/bpf/tc/globals/<map name> */ 30 #define PIN_GLOBAL_NS 2 31 32 /* Must match struct bpf_elf_map layout from iproute2 */ 33 struct { 34 __u32 type; 35 __u32 size_key; 36 __u32 size_value; 37 __u32 max_elem; 38 __u32 flags; 39 __u32 id; 40 __u32 pinning; 41 } server_map SEC("maps") = { 42 .type = BPF_MAP_TYPE_SOCKMAP, 43 .size_key = sizeof(int), 44 .size_value = sizeof(__u64), 45 .max_elem = 1, 46 .pinning = PIN_GLOBAL_NS, 47 }; 48 #endif 49 50 char _license[] SEC("license") = "GPL"; 51 52 /* Fill 'tuple' with L3 info, and attempt to find L4. On fail, return NULL. */ 53 static inline struct bpf_sock_tuple * 54 get_tuple(struct __sk_buff *skb, bool *ipv4, bool *tcp) 55 { 56 void *data_end = (void *)(long)skb->data_end; 57 void *data = (void *)(long)skb->data; 58 struct bpf_sock_tuple *result; 59 struct ethhdr *eth; 60 __u64 tuple_len; 61 __u8 proto = 0; 62 __u64 ihl_len; 63 64 eth = (struct ethhdr *)(data); 65 if (eth + 1 > data_end) 66 return NULL; 67 68 if (eth->h_proto == bpf_htons(ETH_P_IP)) { 69 struct iphdr *iph = (struct iphdr *)(data + sizeof(*eth)); 70 71 if (iph + 1 > data_end) 72 return NULL; 73 if (iph->ihl != 5) 74 /* Options are not supported */ 75 return NULL; 76 ihl_len = iph->ihl * 4; 77 proto = iph->protocol; 78 *ipv4 = true; 79 result = (struct bpf_sock_tuple *)&iph->saddr; 80 } else if (eth->h_proto == bpf_htons(ETH_P_IPV6)) { 81 struct ipv6hdr *ip6h = (struct ipv6hdr *)(data + sizeof(*eth)); 82 83 if (ip6h + 1 > data_end) 84 return NULL; 85 ihl_len = sizeof(*ip6h); 86 proto = ip6h->nexthdr; 87 *ipv4 = false; 88 result = (struct bpf_sock_tuple *)&ip6h->saddr; 89 } else { 90 return (struct bpf_sock_tuple *)data; 91 } 92 93 if (proto != IPPROTO_TCP && proto != IPPROTO_UDP) 94 return NULL; 95 96 *tcp = (proto == IPPROTO_TCP); 97 return result; 98 } 99 100 static inline int 101 handle_udp(struct __sk_buff *skb, struct bpf_sock_tuple *tuple, bool ipv4) 102 { 103 struct bpf_sock *sk; 104 const int zero = 0; 105 size_t tuple_len; 106 __be16 dport; 107 int ret; 108 109 tuple_len = ipv4 ? sizeof(tuple->ipv4) : sizeof(tuple->ipv6); 110 if ((void *)tuple + tuple_len > (void *)(long)skb->data_end) 111 return TC_ACT_SHOT; 112 113 sk = bpf_sk_lookup_udp(skb, tuple, tuple_len, BPF_F_CURRENT_NETNS, 0); 114 if (sk) 115 goto assign; 116 117 dport = ipv4 ? tuple->ipv4.dport : tuple->ipv6.dport; 118 if (dport != bpf_htons(4321)) 119 return TC_ACT_OK; 120 121 sk = bpf_map_lookup_elem(&server_map, &zero); 122 if (!sk) 123 return TC_ACT_SHOT; 124 125 assign: 126 ret = bpf_sk_assign(skb, sk, 0); 127 bpf_sk_release(sk); 128 return ret; 129 } 130 131 static inline int 132 handle_tcp(struct __sk_buff *skb, struct bpf_sock_tuple *tuple, bool ipv4) 133 { 134 struct bpf_sock *sk; 135 const int zero = 0; 136 size_t tuple_len; 137 __be16 dport; 138 int ret; 139 140 tuple_len = ipv4 ? sizeof(tuple->ipv4) : sizeof(tuple->ipv6); 141 if ((void *)tuple + tuple_len > (void *)(long)skb->data_end) 142 return TC_ACT_SHOT; 143 144 sk = bpf_skc_lookup_tcp(skb, tuple, tuple_len, BPF_F_CURRENT_NETNS, 0); 145 if (sk) { 146 if (sk->state != BPF_TCP_LISTEN) 147 goto assign; 148 bpf_sk_release(sk); 149 } 150 151 dport = ipv4 ? tuple->ipv4.dport : tuple->ipv6.dport; 152 if (dport != bpf_htons(4321)) 153 return TC_ACT_OK; 154 155 sk = bpf_map_lookup_elem(&server_map, &zero); 156 if (!sk) 157 return TC_ACT_SHOT; 158 159 if (sk->state != BPF_TCP_LISTEN) { 160 bpf_sk_release(sk); 161 return TC_ACT_SHOT; 162 } 163 164 assign: 165 ret = bpf_sk_assign(skb, sk, 0); 166 bpf_sk_release(sk); 167 return ret; 168 } 169 170 SEC("tc") 171 int bpf_sk_assign_test(struct __sk_buff *skb) 172 { 173 struct bpf_sock_tuple *tuple; 174 bool ipv4 = false; 175 bool tcp = false; 176 int tuple_len; 177 int ret = 0; 178 179 tuple = get_tuple(skb, &ipv4, &tcp); 180 if (!tuple) 181 return TC_ACT_SHOT; 182 183 /* Note that the verifier socket return type for bpf_skc_lookup_tcp() 184 * differs from bpf_sk_lookup_udp(), so even though the C-level type is 185 * the same here, if we try to share the implementations they will 186 * fail to verify because we're crossing pointer types. 187 */ 188 if (tcp) 189 ret = handle_tcp(skb, tuple, ipv4); 190 else 191 ret = handle_udp(skb, tuple, ipv4); 192 193 return ret == 0 ? TC_ACT_OK : TC_ACT_SHOT; 194 } 195