1
2
3#include <stddef.h>
4#include <stdbool.h>
5#include <string.h>
6#include <linux/pkt_cls.h>
7#include <linux/bpf.h>
8#include <linux/in.h>
9#include <linux/if_ether.h>
10#include <linux/ip.h>
11#include <linux/ipv6.h>
12#include <linux/icmp.h>
13#include <linux/icmpv6.h>
14#include <linux/tcp.h>
15#include <linux/udp.h>
16#include "bpf_helpers.h"
17#include "test_iptunnel_common.h"
18#include "bpf_endian.h"
19
20int _version SEC("version") = 1;
21
22static __u32 rol32(__u32 word, unsigned int shift)
23{
24 return (word << shift) | (word >> ((-shift) & 31));
25}
26
27
28
29
30#define __jhash_mix(a, b, c) \
31{ \
32 a -= c; a ^= rol32(c, 4); c += b; \
33 b -= a; b ^= rol32(a, 6); a += c; \
34 c -= b; c ^= rol32(b, 8); b += a; \
35 a -= c; a ^= rol32(c, 16); c += b; \
36 b -= a; b ^= rol32(a, 19); a += c; \
37 c -= b; c ^= rol32(b, 4); b += a; \
38}
39
40#define __jhash_final(a, b, c) \
41{ \
42 c ^= b; c -= rol32(b, 14); \
43 a ^= c; a -= rol32(c, 11); \
44 b ^= a; b -= rol32(a, 25); \
45 c ^= b; c -= rol32(b, 16); \
46 a ^= c; a -= rol32(c, 4); \
47 b ^= a; b -= rol32(a, 14); \
48 c ^= b; c -= rol32(b, 24); \
49}
50
51#define JHASH_INITVAL 0xdeadbeef
52
53typedef unsigned int u32;
54
55static u32 jhash(const void *key, u32 length, u32 initval)
56{
57 u32 a, b, c;
58 const unsigned char *k = key;
59
60 a = b = c = JHASH_INITVAL + length + initval;
61
62 while (length > 12) {
63 a += *(u32 *)(k);
64 b += *(u32 *)(k + 4);
65 c += *(u32 *)(k + 8);
66 __jhash_mix(a, b, c);
67 length -= 12;
68 k += 12;
69 }
70 switch (length) {
71 case 12: c += (u32)k[11]<<24;
72 case 11: c += (u32)k[10]<<16;
73 case 10: c += (u32)k[9]<<8;
74 case 9: c += k[8];
75 case 8: b += (u32)k[7]<<24;
76 case 7: b += (u32)k[6]<<16;
77 case 6: b += (u32)k[5]<<8;
78 case 5: b += k[4];
79 case 4: a += (u32)k[3]<<24;
80 case 3: a += (u32)k[2]<<16;
81 case 2: a += (u32)k[1]<<8;
82 case 1: a += k[0];
83 __jhash_final(a, b, c);
84 case 0:
85 break;
86 }
87
88 return c;
89}
90
91static u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
92{
93 a += initval;
94 b += initval;
95 c += initval;
96 __jhash_final(a, b, c);
97 return c;
98}
99
100static u32 jhash_2words(u32 a, u32 b, u32 initval)
101{
102 return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
103}
104
105#define PCKT_FRAGMENTED 65343
106#define IPV4_HDR_LEN_NO_OPT 20
107#define IPV4_PLUS_ICMP_HDR 28
108#define IPV6_PLUS_ICMP_HDR 48
109#define RING_SIZE 2
110#define MAX_VIPS 12
111#define MAX_REALS 5
112#define CTL_MAP_SIZE 16
113#define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE)
114#define F_IPV6 (1 << 0)
115#define F_HASH_NO_SRC_PORT (1 << 0)
116#define F_ICMP (1 << 0)
117#define F_SYN_SET (1 << 1)
118
119struct packet_description {
120 union {
121 __be32 src;
122 __be32 srcv6[4];
123 };
124 union {
125 __be32 dst;
126 __be32 dstv6[4];
127 };
128 union {
129 __u32 ports;
130 __u16 port16[2];
131 };
132 __u8 proto;
133 __u8 flags;
134};
135
136struct ctl_value {
137 union {
138 __u64 value;
139 __u32 ifindex;
140 __u8 mac[6];
141 };
142};
143
144struct vip_meta {
145 __u32 flags;
146 __u32 vip_num;
147};
148
149struct real_definition {
150 union {
151 __be32 dst;
152 __be32 dstv6[4];
153 };
154 __u8 flags;
155};
156
157struct vip_stats {
158 __u64 bytes;
159 __u64 pkts;
160};
161
162struct eth_hdr {
163 unsigned char eth_dest[ETH_ALEN];
164 unsigned char eth_source[ETH_ALEN];
165 unsigned short eth_proto;
166};
167
168struct bpf_map_def SEC("maps") vip_map = {
169 .type = BPF_MAP_TYPE_HASH,
170 .key_size = sizeof(struct vip),
171 .value_size = sizeof(struct vip_meta),
172 .max_entries = MAX_VIPS,
173};
174
175struct bpf_map_def SEC("maps") ch_rings = {
176 .type = BPF_MAP_TYPE_ARRAY,
177 .key_size = sizeof(__u32),
178 .value_size = sizeof(__u32),
179 .max_entries = CH_RINGS_SIZE,
180};
181
182struct bpf_map_def SEC("maps") reals = {
183 .type = BPF_MAP_TYPE_ARRAY,
184 .key_size = sizeof(__u32),
185 .value_size = sizeof(struct real_definition),
186 .max_entries = MAX_REALS,
187};
188
189struct bpf_map_def SEC("maps") stats = {
190 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
191 .key_size = sizeof(__u32),
192 .value_size = sizeof(struct vip_stats),
193 .max_entries = MAX_VIPS,
194};
195
196struct bpf_map_def SEC("maps") ctl_array = {
197 .type = BPF_MAP_TYPE_ARRAY,
198 .key_size = sizeof(__u32),
199 .value_size = sizeof(struct ctl_value),
200 .max_entries = CTL_MAP_SIZE,
201};
202
203static __u32 get_packet_hash(struct packet_description *pckt,
204 bool ipv6)
205{
206 if (ipv6)
207 return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
208 pckt->ports, CH_RINGS_SIZE);
209 else
210 return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
211}
212
213static bool get_packet_dst(struct real_definition **real,
214 struct packet_description *pckt,
215 struct vip_meta *vip_info,
216 bool is_ipv6)
217{
218 __u32 hash = get_packet_hash(pckt, is_ipv6);
219 __u32 key = RING_SIZE * vip_info->vip_num + hash % RING_SIZE;
220 __u32 *real_pos;
221
222 if (hash != 0x358459b7 &&
223 hash != 0x2f4bc6bb )
224 return 0;
225
226 real_pos = bpf_map_lookup_elem(&ch_rings, &key);
227 if (!real_pos)
228 return false;
229 key = *real_pos;
230 *real = bpf_map_lookup_elem(&reals, &key);
231 if (!(*real))
232 return false;
233 return true;
234}
235
236static int parse_icmpv6(void *data, void *data_end, __u64 off,
237 struct packet_description *pckt)
238{
239 struct icmp6hdr *icmp_hdr;
240 struct ipv6hdr *ip6h;
241
242 icmp_hdr = data + off;
243 if (icmp_hdr + 1 > data_end)
244 return TC_ACT_SHOT;
245 if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
246 return TC_ACT_OK;
247 off += sizeof(struct icmp6hdr);
248 ip6h = data + off;
249 if (ip6h + 1 > data_end)
250 return TC_ACT_SHOT;
251 pckt->proto = ip6h->nexthdr;
252 pckt->flags |= F_ICMP;
253 memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
254 memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
255 return TC_ACT_UNSPEC;
256}
257
258static int parse_icmp(void *data, void *data_end, __u64 off,
259 struct packet_description *pckt)
260{
261 struct icmphdr *icmp_hdr;
262 struct iphdr *iph;
263
264 icmp_hdr = data + off;
265 if (icmp_hdr + 1 > data_end)
266 return TC_ACT_SHOT;
267 if (icmp_hdr->type != ICMP_DEST_UNREACH ||
268 icmp_hdr->code != ICMP_FRAG_NEEDED)
269 return TC_ACT_OK;
270 off += sizeof(struct icmphdr);
271 iph = data + off;
272 if (iph + 1 > data_end)
273 return TC_ACT_SHOT;
274 if (iph->ihl != 5)
275 return TC_ACT_SHOT;
276 pckt->proto = iph->protocol;
277 pckt->flags |= F_ICMP;
278 pckt->src = iph->daddr;
279 pckt->dst = iph->saddr;
280 return TC_ACT_UNSPEC;
281}
282
283static bool parse_udp(void *data, __u64 off, void *data_end,
284 struct packet_description *pckt)
285{
286 struct udphdr *udp;
287 udp = data + off;
288
289 if (udp + 1 > data_end)
290 return false;
291
292 if (!(pckt->flags & F_ICMP)) {
293 pckt->port16[0] = udp->source;
294 pckt->port16[1] = udp->dest;
295 } else {
296 pckt->port16[0] = udp->dest;
297 pckt->port16[1] = udp->source;
298 }
299 return true;
300}
301
302static bool parse_tcp(void *data, __u64 off, void *data_end,
303 struct packet_description *pckt)
304{
305 struct tcphdr *tcp;
306
307 tcp = data + off;
308 if (tcp + 1 > data_end)
309 return false;
310
311 if (tcp->syn)
312 pckt->flags |= F_SYN_SET;
313
314 if (!(pckt->flags & F_ICMP)) {
315 pckt->port16[0] = tcp->source;
316 pckt->port16[1] = tcp->dest;
317 } else {
318 pckt->port16[0] = tcp->dest;
319 pckt->port16[1] = tcp->source;
320 }
321 return true;
322}
323
324static int process_packet(void *data, __u64 off, void *data_end,
325 bool is_ipv6, struct __sk_buff *skb)
326{
327 void *pkt_start = (void *)(long)skb->data;
328 struct packet_description pckt = {};
329 struct eth_hdr *eth = pkt_start;
330 struct bpf_tunnel_key tkey = {};
331 struct vip_stats *data_stats;
332 struct real_definition *dst;
333 struct vip_meta *vip_info;
334 struct ctl_value *cval;
335 __u32 v4_intf_pos = 1;
336 __u32 v6_intf_pos = 2;
337 struct ipv6hdr *ip6h;
338 struct vip vip = {};
339 struct iphdr *iph;
340 int tun_flag = 0;
341 __u16 pkt_bytes;
342 __u64 iph_len;
343 __u32 ifindex;
344 __u8 protocol;
345 __u32 vip_num;
346 int action;
347
348 tkey.tunnel_ttl = 64;
349 if (is_ipv6) {
350 ip6h = data + off;
351 if (ip6h + 1 > data_end)
352 return TC_ACT_SHOT;
353
354 iph_len = sizeof(struct ipv6hdr);
355 protocol = ip6h->nexthdr;
356 pckt.proto = protocol;
357 pkt_bytes = bpf_ntohs(ip6h->payload_len);
358 off += iph_len;
359 if (protocol == IPPROTO_FRAGMENT) {
360 return TC_ACT_SHOT;
361 } else if (protocol == IPPROTO_ICMPV6) {
362 action = parse_icmpv6(data, data_end, off, &pckt);
363 if (action >= 0)
364 return action;
365 off += IPV6_PLUS_ICMP_HDR;
366 } else {
367 memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
368 memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
369 }
370 } else {
371 iph = data + off;
372 if (iph + 1 > data_end)
373 return TC_ACT_SHOT;
374 if (iph->ihl != 5)
375 return TC_ACT_SHOT;
376
377 protocol = iph->protocol;
378 pckt.proto = protocol;
379 pkt_bytes = bpf_ntohs(iph->tot_len);
380 off += IPV4_HDR_LEN_NO_OPT;
381
382 if (iph->frag_off & PCKT_FRAGMENTED)
383 return TC_ACT_SHOT;
384 if (protocol == IPPROTO_ICMP) {
385 action = parse_icmp(data, data_end, off, &pckt);
386 if (action >= 0)
387 return action;
388 off += IPV4_PLUS_ICMP_HDR;
389 } else {
390 pckt.src = iph->saddr;
391 pckt.dst = iph->daddr;
392 }
393 }
394 protocol = pckt.proto;
395
396 if (protocol == IPPROTO_TCP) {
397 if (!parse_tcp(data, off, data_end, &pckt))
398 return TC_ACT_SHOT;
399 } else if (protocol == IPPROTO_UDP) {
400 if (!parse_udp(data, off, data_end, &pckt))
401 return TC_ACT_SHOT;
402 } else {
403 return TC_ACT_SHOT;
404 }
405
406 if (is_ipv6)
407 memcpy(vip.daddr.v6, pckt.dstv6, 16);
408 else
409 vip.daddr.v4 = pckt.dst;
410
411 vip.dport = pckt.port16[1];
412 vip.protocol = pckt.proto;
413 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
414 if (!vip_info) {
415 vip.dport = 0;
416 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
417 if (!vip_info)
418 return TC_ACT_SHOT;
419 pckt.port16[1] = 0;
420 }
421
422 if (vip_info->flags & F_HASH_NO_SRC_PORT)
423 pckt.port16[0] = 0;
424
425 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
426 return TC_ACT_SHOT;
427
428 if (dst->flags & F_IPV6) {
429 cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
430 if (!cval)
431 return TC_ACT_SHOT;
432 ifindex = cval->ifindex;
433 memcpy(tkey.remote_ipv6, dst->dstv6, 16);
434 tun_flag = BPF_F_TUNINFO_IPV6;
435 } else {
436 cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
437 if (!cval)
438 return TC_ACT_SHOT;
439 ifindex = cval->ifindex;
440 tkey.remote_ipv4 = dst->dst;
441 }
442 vip_num = vip_info->vip_num;
443 data_stats = bpf_map_lookup_elem(&stats, &vip_num);
444 if (!data_stats)
445 return TC_ACT_SHOT;
446 data_stats->pkts++;
447 data_stats->bytes += pkt_bytes;
448 bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
449 *(u32 *)eth->eth_dest = tkey.remote_ipv4;
450 return bpf_redirect(ifindex, 0);
451}
452
453SEC("l4lb-demo")
454int balancer_ingress(struct __sk_buff *ctx)
455{
456 void *data_end = (void *)(long)ctx->data_end;
457 void *data = (void *)(long)ctx->data;
458 struct eth_hdr *eth = data;
459 __u32 eth_proto;
460 __u32 nh_off;
461
462 nh_off = sizeof(struct eth_hdr);
463 if (data + nh_off > data_end)
464 return TC_ACT_SHOT;
465 eth_proto = eth->eth_proto;
466 if (eth_proto == bpf_htons(ETH_P_IP))
467 return process_packet(data, nh_off, data_end, false, ctx);
468 else if (eth_proto == bpf_htons(ETH_P_IPV6))
469 return process_packet(data, nh_off, data_end, true, ctx);
470 else
471 return TC_ACT_SHOT;
472}
473char _license[] SEC("license") = "GPL";
474