Loading...
1// SPDX-License-Identifier: GPL-2.0
2// Copyright (c) 2017 Facebook
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/bpf_helpers.h>
17#include "test_iptunnel_common.h"
18#include <bpf/bpf_endian.h>
19
20static __always_inline __u32 rol32(__u32 word, unsigned int shift)
21{
22 return (word << shift) | (word >> ((-shift) & 31));
23}
24
25/* copy paste of jhash from kernel sources to make sure llvm
26 * can compile it into valid sequence of bpf instructions
27 */
28#define __jhash_mix(a, b, c) \
29{ \
30 a -= c; a ^= rol32(c, 4); c += b; \
31 b -= a; b ^= rol32(a, 6); a += c; \
32 c -= b; c ^= rol32(b, 8); b += a; \
33 a -= c; a ^= rol32(c, 16); c += b; \
34 b -= a; b ^= rol32(a, 19); a += c; \
35 c -= b; c ^= rol32(b, 4); b += a; \
36}
37
38#define __jhash_final(a, b, c) \
39{ \
40 c ^= b; c -= rol32(b, 14); \
41 a ^= c; a -= rol32(c, 11); \
42 b ^= a; b -= rol32(a, 25); \
43 c ^= b; c -= rol32(b, 16); \
44 a ^= c; a -= rol32(c, 4); \
45 b ^= a; b -= rol32(a, 14); \
46 c ^= b; c -= rol32(b, 24); \
47}
48
49#define JHASH_INITVAL 0xdeadbeef
50
51typedef unsigned int u32;
52
53static __noinline u32 jhash(const void *key, u32 length, u32 initval)
54{
55 u32 a, b, c;
56 const unsigned char *k = key;
57
58 a = b = c = JHASH_INITVAL + length + initval;
59
60 while (length > 12) {
61 a += *(u32 *)(k);
62 b += *(u32 *)(k + 4);
63 c += *(u32 *)(k + 8);
64 __jhash_mix(a, b, c);
65 length -= 12;
66 k += 12;
67 }
68 switch (length) {
69 case 12: c += (u32)k[11]<<24;
70 case 11: c += (u32)k[10]<<16;
71 case 10: c += (u32)k[9]<<8;
72 case 9: c += k[8];
73 case 8: b += (u32)k[7]<<24;
74 case 7: b += (u32)k[6]<<16;
75 case 6: b += (u32)k[5]<<8;
76 case 5: b += k[4];
77 case 4: a += (u32)k[3]<<24;
78 case 3: a += (u32)k[2]<<16;
79 case 2: a += (u32)k[1]<<8;
80 case 1: a += k[0];
81 __jhash_final(a, b, c);
82 case 0: /* Nothing left to add */
83 break;
84 }
85
86 return c;
87}
88
89static __noinline u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
90{
91 a += initval;
92 b += initval;
93 c += initval;
94 __jhash_final(a, b, c);
95 return c;
96}
97
98static __noinline u32 jhash_2words(u32 a, u32 b, u32 initval)
99{
100 return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
101}
102
103#define PCKT_FRAGMENTED 65343
104#define IPV4_HDR_LEN_NO_OPT 20
105#define IPV4_PLUS_ICMP_HDR 28
106#define IPV6_PLUS_ICMP_HDR 48
107#define RING_SIZE 2
108#define MAX_VIPS 12
109#define MAX_REALS 5
110#define CTL_MAP_SIZE 16
111#define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE)
112#define F_IPV6 (1 << 0)
113#define F_HASH_NO_SRC_PORT (1 << 0)
114#define F_ICMP (1 << 0)
115#define F_SYN_SET (1 << 1)
116
117struct packet_description {
118 union {
119 __be32 src;
120 __be32 srcv6[4];
121 };
122 union {
123 __be32 dst;
124 __be32 dstv6[4];
125 };
126 union {
127 __u32 ports;
128 __u16 port16[2];
129 };
130 __u8 proto;
131 __u8 flags;
132};
133
134struct ctl_value {
135 union {
136 __u64 value;
137 __u32 ifindex;
138 __u8 mac[6];
139 };
140};
141
142struct vip_meta {
143 __u32 flags;
144 __u32 vip_num;
145};
146
147struct real_definition {
148 union {
149 __be32 dst;
150 __be32 dstv6[4];
151 };
152 __u8 flags;
153};
154
155struct vip_stats {
156 __u64 bytes;
157 __u64 pkts;
158};
159
160struct eth_hdr {
161 unsigned char eth_dest[ETH_ALEN];
162 unsigned char eth_source[ETH_ALEN];
163 unsigned short eth_proto;
164};
165
166struct {
167 __uint(type, BPF_MAP_TYPE_HASH);
168 __uint(max_entries, MAX_VIPS);
169 __type(key, struct vip);
170 __type(value, struct vip_meta);
171} vip_map SEC(".maps");
172
173struct {
174 __uint(type, BPF_MAP_TYPE_ARRAY);
175 __uint(max_entries, CH_RINGS_SIZE);
176 __type(key, __u32);
177 __type(value, __u32);
178} ch_rings SEC(".maps");
179
180struct {
181 __uint(type, BPF_MAP_TYPE_ARRAY);
182 __uint(max_entries, MAX_REALS);
183 __type(key, __u32);
184 __type(value, struct real_definition);
185} reals SEC(".maps");
186
187struct {
188 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
189 __uint(max_entries, MAX_VIPS);
190 __type(key, __u32);
191 __type(value, struct vip_stats);
192} stats SEC(".maps");
193
194struct {
195 __uint(type, BPF_MAP_TYPE_ARRAY);
196 __uint(max_entries, CTL_MAP_SIZE);
197 __type(key, __u32);
198 __type(value, struct ctl_value);
199} ctl_array SEC(".maps");
200
201static __noinline __u32 get_packet_hash(struct packet_description *pckt, bool ipv6)
202{
203 if (ipv6)
204 return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
205 pckt->ports, CH_RINGS_SIZE);
206 else
207 return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
208}
209
210static __noinline bool get_packet_dst(struct real_definition **real,
211 struct packet_description *pckt,
212 struct vip_meta *vip_info,
213 bool is_ipv6)
214{
215 __u32 hash = get_packet_hash(pckt, is_ipv6);
216 __u32 key = RING_SIZE * vip_info->vip_num + hash % RING_SIZE;
217 __u32 *real_pos;
218
219 if (hash != 0x358459b7 /* jhash of ipv4 packet */ &&
220 hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
221 return false;
222
223 real_pos = bpf_map_lookup_elem(&ch_rings, &key);
224 if (!real_pos)
225 return false;
226 key = *real_pos;
227 *real = bpf_map_lookup_elem(&reals, &key);
228 if (!(*real))
229 return false;
230 return true;
231}
232
233static __noinline int parse_icmpv6(void *data, void *data_end, __u64 off,
234 struct packet_description *pckt)
235{
236 struct icmp6hdr *icmp_hdr;
237 struct ipv6hdr *ip6h;
238
239 icmp_hdr = data + off;
240 if (icmp_hdr + 1 > data_end)
241 return TC_ACT_SHOT;
242 if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
243 return TC_ACT_OK;
244 off += sizeof(struct icmp6hdr);
245 ip6h = data + off;
246 if (ip6h + 1 > data_end)
247 return TC_ACT_SHOT;
248 pckt->proto = ip6h->nexthdr;
249 pckt->flags |= F_ICMP;
250 memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
251 memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
252 return TC_ACT_UNSPEC;
253}
254
255static __noinline int parse_icmp(void *data, void *data_end, __u64 off,
256 struct packet_description *pckt)
257{
258 struct icmphdr *icmp_hdr;
259 struct iphdr *iph;
260
261 icmp_hdr = data + off;
262 if (icmp_hdr + 1 > data_end)
263 return TC_ACT_SHOT;
264 if (icmp_hdr->type != ICMP_DEST_UNREACH ||
265 icmp_hdr->code != ICMP_FRAG_NEEDED)
266 return TC_ACT_OK;
267 off += sizeof(struct icmphdr);
268 iph = data + off;
269 if (iph + 1 > data_end)
270 return TC_ACT_SHOT;
271 if (iph->ihl != 5)
272 return TC_ACT_SHOT;
273 pckt->proto = iph->protocol;
274 pckt->flags |= F_ICMP;
275 pckt->src = iph->daddr;
276 pckt->dst = iph->saddr;
277 return TC_ACT_UNSPEC;
278}
279
280static __noinline bool parse_udp(void *data, __u64 off, void *data_end,
281 struct packet_description *pckt)
282{
283 struct udphdr *udp;
284 udp = data + off;
285
286 if (udp + 1 > data_end)
287 return false;
288
289 if (!(pckt->flags & F_ICMP)) {
290 pckt->port16[0] = udp->source;
291 pckt->port16[1] = udp->dest;
292 } else {
293 pckt->port16[0] = udp->dest;
294 pckt->port16[1] = udp->source;
295 }
296 return true;
297}
298
299static __noinline bool parse_tcp(void *data, __u64 off, void *data_end,
300 struct packet_description *pckt)
301{
302 struct tcphdr *tcp;
303
304 tcp = data + off;
305 if (tcp + 1 > data_end)
306 return false;
307
308 if (tcp->syn)
309 pckt->flags |= F_SYN_SET;
310
311 if (!(pckt->flags & F_ICMP)) {
312 pckt->port16[0] = tcp->source;
313 pckt->port16[1] = tcp->dest;
314 } else {
315 pckt->port16[0] = tcp->dest;
316 pckt->port16[1] = tcp->source;
317 }
318 return true;
319}
320
321static __noinline int process_packet(void *data, __u64 off, void *data_end,
322 bool is_ipv6, struct __sk_buff *skb)
323{
324 void *pkt_start = (void *)(long)skb->data;
325 struct packet_description pckt = {};
326 struct eth_hdr *eth = pkt_start;
327 struct bpf_tunnel_key tkey = {};
328 struct vip_stats *data_stats;
329 struct real_definition *dst;
330 struct vip_meta *vip_info;
331 struct ctl_value *cval;
332 __u32 v4_intf_pos = 1;
333 __u32 v6_intf_pos = 2;
334 struct ipv6hdr *ip6h;
335 struct vip vip = {};
336 struct iphdr *iph;
337 int tun_flag = 0;
338 __u16 pkt_bytes;
339 __u64 iph_len;
340 __u32 ifindex;
341 __u8 protocol;
342 __u32 vip_num;
343 int action;
344
345 tkey.tunnel_ttl = 64;
346 if (is_ipv6) {
347 ip6h = data + off;
348 if (ip6h + 1 > data_end)
349 return TC_ACT_SHOT;
350
351 iph_len = sizeof(struct ipv6hdr);
352 protocol = ip6h->nexthdr;
353 pckt.proto = protocol;
354 pkt_bytes = bpf_ntohs(ip6h->payload_len);
355 off += iph_len;
356 if (protocol == IPPROTO_FRAGMENT) {
357 return TC_ACT_SHOT;
358 } else if (protocol == IPPROTO_ICMPV6) {
359 action = parse_icmpv6(data, data_end, off, &pckt);
360 if (action >= 0)
361 return action;
362 off += IPV6_PLUS_ICMP_HDR;
363 } else {
364 memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
365 memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
366 }
367 } else {
368 iph = data + off;
369 if (iph + 1 > data_end)
370 return TC_ACT_SHOT;
371 if (iph->ihl != 5)
372 return TC_ACT_SHOT;
373
374 protocol = iph->protocol;
375 pckt.proto = protocol;
376 pkt_bytes = bpf_ntohs(iph->tot_len);
377 off += IPV4_HDR_LEN_NO_OPT;
378
379 if (iph->frag_off & PCKT_FRAGMENTED)
380 return TC_ACT_SHOT;
381 if (protocol == IPPROTO_ICMP) {
382 action = parse_icmp(data, data_end, off, &pckt);
383 if (action >= 0)
384 return action;
385 off += IPV4_PLUS_ICMP_HDR;
386 } else {
387 pckt.src = iph->saddr;
388 pckt.dst = iph->daddr;
389 }
390 }
391 protocol = pckt.proto;
392
393 if (protocol == IPPROTO_TCP) {
394 if (!parse_tcp(data, off, data_end, &pckt))
395 return TC_ACT_SHOT;
396 } else if (protocol == IPPROTO_UDP) {
397 if (!parse_udp(data, off, data_end, &pckt))
398 return TC_ACT_SHOT;
399 } else {
400 return TC_ACT_SHOT;
401 }
402
403 if (is_ipv6)
404 memcpy(vip.daddr.v6, pckt.dstv6, 16);
405 else
406 vip.daddr.v4 = pckt.dst;
407
408 vip.dport = pckt.port16[1];
409 vip.protocol = pckt.proto;
410 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
411 if (!vip_info) {
412 vip.dport = 0;
413 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
414 if (!vip_info)
415 return TC_ACT_SHOT;
416 pckt.port16[1] = 0;
417 }
418
419 if (vip_info->flags & F_HASH_NO_SRC_PORT)
420 pckt.port16[0] = 0;
421
422 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
423 return TC_ACT_SHOT;
424
425 if (dst->flags & F_IPV6) {
426 cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
427 if (!cval)
428 return TC_ACT_SHOT;
429 ifindex = cval->ifindex;
430 memcpy(tkey.remote_ipv6, dst->dstv6, 16);
431 tun_flag = BPF_F_TUNINFO_IPV6;
432 } else {
433 cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
434 if (!cval)
435 return TC_ACT_SHOT;
436 ifindex = cval->ifindex;
437 tkey.remote_ipv4 = dst->dst;
438 }
439 vip_num = vip_info->vip_num;
440 data_stats = bpf_map_lookup_elem(&stats, &vip_num);
441 if (!data_stats)
442 return TC_ACT_SHOT;
443 data_stats->pkts++;
444 data_stats->bytes += pkt_bytes;
445 bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
446 *(u32 *)eth->eth_dest = tkey.remote_ipv4;
447 return bpf_redirect(ifindex, 0);
448}
449
450SEC("tc")
451int balancer_ingress(struct __sk_buff *ctx)
452{
453 void *data_end = (void *)(long)ctx->data_end;
454 void *data = (void *)(long)ctx->data;
455 struct eth_hdr *eth = data;
456 __u32 eth_proto;
457 __u32 nh_off;
458
459 nh_off = sizeof(struct eth_hdr);
460 if (data + nh_off > data_end)
461 return TC_ACT_SHOT;
462 eth_proto = eth->eth_proto;
463 if (eth_proto == bpf_htons(ETH_P_IP))
464 return process_packet(data, nh_off, data_end, false, ctx);
465 else if (eth_proto == bpf_htons(ETH_P_IPV6))
466 return process_packet(data, nh_off, data_end, true, ctx);
467 else
468 return TC_ACT_SHOT;
469}
470char _license[] SEC("license") = "GPL";
1// SPDX-License-Identifier: GPL-2.0
2// Copyright (c) 2017 Facebook
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/bpf_helpers.h>
17#include "test_iptunnel_common.h"
18#include <bpf/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/* copy paste of jhash from kernel sources to make sure llvm
28 * can compile it into valid sequence of bpf instructions
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: /* Nothing left to add */
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 {
169 __uint(type, BPF_MAP_TYPE_HASH);
170 __uint(max_entries, MAX_VIPS);
171 __type(key, struct vip);
172 __type(value, struct vip_meta);
173} vip_map SEC(".maps");
174
175struct {
176 __uint(type, BPF_MAP_TYPE_ARRAY);
177 __uint(max_entries, CH_RINGS_SIZE);
178 __type(key, __u32);
179 __type(value, __u32);
180} ch_rings SEC(".maps");
181
182struct {
183 __uint(type, BPF_MAP_TYPE_ARRAY);
184 __uint(max_entries, MAX_REALS);
185 __type(key, __u32);
186 __type(value, struct real_definition);
187} reals SEC(".maps");
188
189struct {
190 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
191 __uint(max_entries, MAX_VIPS);
192 __type(key, __u32);
193 __type(value, struct vip_stats);
194} stats SEC(".maps");
195
196struct {
197 __uint(type, BPF_MAP_TYPE_ARRAY);
198 __uint(max_entries, CTL_MAP_SIZE);
199 __type(key, __u32);
200 __type(value, struct ctl_value);
201} ctl_array SEC(".maps");
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 /* jhash of ipv4 packet */ &&
223 hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
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";