3 #include <linux/if_ether.h>
6 #include <linux/icmp.h>
7 #include <linux/icmpv6.h>
13 #define IP_CE 0x8000 /* Flag: "Congestion" */
14 #define IP_DF 0x4000 /* Flag: "Don't Fragment" */
15 #define IP_MF 0x2000 /* Flag: "More Fragments" */
16 #define IP_OFFSET 0x1FFF /* "Fragment Offset" part */
18 #define IP_PROTO_TCP 6
19 #define IP_PROTO_UDP 17
20 #define IP_PROTO_ICMP 1
21 #define IP6_PROTO_ICMPV6 58
22 #define IP6_PROTO_FRAG 44
24 typedef __uint128_t uint128_t;
26 // Our own ipv6hdr that uses uint128_t
28 #if defined(__LITTLE_ENDIAN_BITFIELD)
31 #elif defined(__BIG_ENDIAN_BITFIELD)
35 #error "Please fix <asm/byteorder.h>"
45 } __attribute__((packed));
48 #define IP6_FRAGOFF 0xfff8
52 uint16_t frag_off; // BE low 3 bits flags, last is "more frags"
54 } __attribute__((packed));
56 // Our own ethhdr with optional vlan tags
58 unsigned char h_dest[ETH_ALEN]; /* destination eth addr */
59 unsigned char h_source[ETH_ALEN]; /* source ether addr */
60 __be16 vlan_magic; /* 0x8100 */
61 __be16 tci; /* PCP (3 bits), DEI (1 bit), and VLAN (12 bits) */
62 __be16 h_proto; /* packet type ID field */
63 } __attribute__((packed));
65 // Our own tcphdr without the flags blown up
75 } __attribute__((packed));
77 // Note that all operations on uint128s *stay* in Network byte order!
79 #if defined(__LITTLE_ENDIAN)
80 #define BIGEND32(v) ((v >> 3*8) | ((v >> 8) & 0xff00) | ((v << 8) & 0xff0000) | (v << 3*8) & 0xff000000)
81 #elif defined(__BIG_ENDIAN)
82 #define BIGEND32(v) (v)
84 #error "Need endian info"
87 #if defined(__LITTLE_ENDIAN)
88 #define BIGEND128(a, b, c, d) ( \
89 (((uint128_t)BIGEND32(d)) << 3*32) | \
90 (((uint128_t)BIGEND32(c)) << 2*32) | \
91 (((uint128_t)BIGEND32(b)) << 1*32) | \
92 (((uint128_t)BIGEND32(a)) << 0*32))
93 #define HTON128(a) BIGEND128(a >> 3*32, a >> 2*32, a >> 1*32, a>> 0*32)
94 // Yes, somehow macro'ing this changes LLVM's view of htons...
95 #define BE16(a) ((((uint16_t)(a & 0xff00)) >> 8) | (((uint16_t)(a & 0xff)) << 8))
96 #elif defined(__BIG_ENDIAN)
97 #define BIGEND128(a, b, c, d) ((((uint128_t)a) << 3*32) | (((uint128_t)b) << 2*32) | (((uint128_t)c) << 1*32) | (((uint128_t)d) << 0*32))
98 #define HTON128(a) (a)
100 #error "Need endian info"
103 #define MASK4(pfxlen) BIGEND32(~((((uint32_t)1) << (32 - pfxlen)) - 1))
104 #define MASK6(pfxlen) HTON128(~((((uint128_t)1) << (128 - pfxlen)) - 1))
105 #define MASK6_OFFS(offs, pfxlen) HTON128((~((((uint128_t)1) << (128 - pfxlen)) - 1)) & ((((uint128_t)1) << (128 - offs)) - 1))
107 // PARSE is used as a preprocessor flag to indicate parsing fields
111 #define unlikely(a) __builtin_expect(a, 0)
112 #define likely(a) __builtin_expect(a, 1)
114 static const uint32_t PKT_LEN_DROP = 0;
115 static const uint32_t VLAN_DROP = 1;
116 static const uint32_t IHL_DROP = 2;
117 static const uint32_t V6FRAG_DROP = 3;
118 #define STATIC_RULE_CNT 4
121 // 64 bit version of xdp_md for testing
126 /* Below access go through struct xdp_rxq_info */
127 __u64 ingress_ifindex; /* rxq->dev->ifindex */
128 __u64 rx_queue_index; /* rxq->queue_index */
130 __u64 egress_ifindex; /* txq->dev->ifindex */
132 static const int XDP_PASS = 0;
133 static const int XDP_DROP = 1;
135 static long drop_cnt_map[RULECNT + STATIC_RULE_CNT];
136 #define DO_RETURN(reason, ret) { \
137 if (ret == XDP_DROP) drop_cnt_map[reason] += 1; \
142 #include <linux/bpf.h>
143 #include <bpf/bpf_helpers.h>
145 struct bpf_map_def SEC("maps") drop_cnt_map = {
146 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
147 .key_size = sizeof(uint32_t),
148 .value_size = sizeof(long),
149 .max_entries = RULECNT + STATIC_RULE_CNT,
151 #define DO_RETURN(reason, ret) {\
152 if (ret == XDP_DROP) { \
153 long *value = bpf_map_lookup_elem(&drop_cnt_map, &reason); \
162 int xdp_drop_prog(struct xdp_md *ctx)
164 const void *const data_end = (void *)(size_t)ctx->data_end;
166 const void * pktdata;
167 unsigned short eth_proto;
170 if (unlikely((void*)(size_t)ctx->data + sizeof(struct ethhdr) > data_end))
171 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
172 const struct ethhdr *const eth = (void*)(size_t)ctx->data;
174 #if PARSE_8021Q == PARSE
175 if (likely(eth->h_proto == BE16(ETH_P_8021Q))) {
176 if (unlikely((void*)(size_t)ctx->data + sizeof(struct ethhdr_vlan) > data_end))
177 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
178 const struct ethhdr_vlan *const eth_vlan = (void*)(size_t)ctx->data;
181 if (unlikely((eth_vlan->tci & BE16(0xfff)) != BE16(REQ_8021Q)))
182 DO_RETURN(VLAN_DROP, XDP_DROP);
185 eth_proto = eth_vlan->h_proto;
186 pktdata = (const void *)(long)ctx->data + sizeof(struct ethhdr_vlan);
188 if (unlikely(eth->h_proto == BE16(ETH_P_8021Q))) {
189 DO_RETURN(VLAN_DROP, PARSE_8021Q);
193 DO_RETURN(VLAN_DROP, XDP_DROP);
195 pktdata = (const void *)(long)ctx->data + sizeof(struct ethhdr);
196 eth_proto = eth->h_proto;
202 const struct iphdr *ip = NULL;
203 const struct icmphdr *icmp = NULL;
206 const struct ip6hdr *ip6 = NULL;
207 const struct icmp6hdr *icmpv6 = NULL;
208 const struct ip6_fraghdr *frag6 = NULL;
211 const void *l4hdr = NULL;
212 const struct tcphdr *tcp = NULL;
213 const struct udphdr *udp = NULL;
216 if (eth_proto == BE16(ETH_P_IP)) {
217 if (unlikely(pktdata + sizeof(struct iphdr) > data_end))
218 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
219 ip = (struct iphdr*) pktdata;
221 #if PARSE_IHL == PARSE
222 if (unlikely(ip->ihl < 5)) DO_RETURN(IHL_DROP, XDP_DROP);
223 l4hdr = pktdata + ip->ihl * 4;
225 if (ip->ihl != 5) DO_RETURN(IHL_DROP, PARSE_IHL);
226 l4hdr = pktdata + 5*4;
229 if ((ip->frag_off & BE16(IP_OFFSET)) == 0) {
230 if (ip->protocol == IP_PROTO_TCP) {
231 if (unlikely(l4hdr + sizeof(struct tcphdr) > data_end))
232 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
233 tcp = (struct tcphdr*) l4hdr;
234 } else if (ip->protocol == IP_PROTO_UDP) {
235 if (unlikely(l4hdr + sizeof(struct udphdr) > data_end))
236 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
237 udp = (struct udphdr*) l4hdr;
238 } else if (ip->protocol == IP_PROTO_ICMP) {
239 if (unlikely(l4hdr + sizeof(struct icmphdr) > data_end))
240 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
241 icmp = (struct icmphdr*) l4hdr;
247 if (eth_proto == BE16(ETH_P_IPV6)) {
248 if (unlikely(pktdata + sizeof(struct ip6hdr) > data_end))
249 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
250 ip6 = (struct ip6hdr*) pktdata;
252 l4hdr = pktdata + 40;
254 uint8_t v6nexthdr = ip6->nexthdr;
256 #if PARSE_V6_FRAG == PARSE
257 if (ip6->nexthdr == IP6_PROTO_FRAG) {
258 if (unlikely(l4hdr + sizeof(struct ip6_fraghdr) > data_end))
259 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
260 frag6 = (struct ip6_fraghdr*) l4hdr;
261 l4hdr = l4hdr + sizeof(struct ip6_fraghdr);
262 v6nexthdr = frag6->nexthdr;
264 if (unlikely(ip6->nexthdr == IP6_PROTO_FRAG)) {
265 DO_RETURN(V6FRAG_DROP, PARSE_V6_FRAG);
269 // TODO: Handle more options?
271 if (frag6 == NULL || (frag6->frag_off & BE16(IP6_FRAGOFF)) == 0) {
272 if (v6nexthdr == IP_PROTO_TCP) {
273 if (unlikely(l4hdr + sizeof(struct tcphdr) > data_end))
274 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
275 tcp = (struct tcphdr*) l4hdr;
276 } else if (v6nexthdr == IP_PROTO_UDP) {
277 if (unlikely(l4hdr + sizeof(struct udphdr) > data_end))
278 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
279 udp = (struct udphdr*) l4hdr;
280 } else if (v6nexthdr == IP6_PROTO_ICMPV6) {
281 if (unlikely(l4hdr + sizeof(struct icmp6hdr) > data_end))
282 DO_RETURN(PKT_LEN_DROP, XDP_DROP);
283 icmpv6 = (struct icmp6hdr*) l4hdr;
289 uint16_t sport, dport; // Host Endian! Only valid with tcp || udp
291 sport = BE16(tcp->source);
292 dport = BE16(tcp->dest);
293 } else if (udp != NULL) {
294 sport = BE16(udp->source);
295 dport = BE16(udp->dest);
307 const char d[] = TEST;
309 struct xdp_md test = {
311 // -1 because sizeof includes a trailing null in the "string"
312 .data_end = (uint64_t)(d + sizeof(d) - 1),
314 assert(xdp_drop_prog(&test) == TEST_EXP);