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
20 #include "bpf_kfuncs.h"
21
rol32(__u32 word,unsigned int shift)22 static __always_inline __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
53 typedef unsigned int u32;
54
jhash(const void * key,u32 length,u32 initval)55 static __noinline 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
__jhash_nwords(u32 a,u32 b,u32 c,u32 initval)91 static __noinline 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
jhash_2words(u32 a,u32 b,u32 initval)100 static __noinline 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
119 struct 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
136 struct ctl_value {
137 union {
138 __u64 value;
139 __u32 ifindex;
140 __u8 mac[6];
141 };
142 };
143
144 struct vip_meta {
145 __u32 flags;
146 __u32 vip_num;
147 };
148
149 struct real_definition {
150 union {
151 __be32 dst;
152 __be32 dstv6[4];
153 };
154 __u8 flags;
155 };
156
157 struct vip_stats {
158 __u64 bytes;
159 __u64 pkts;
160 };
161
162 struct eth_hdr {
163 unsigned char eth_dest[ETH_ALEN];
164 unsigned char eth_source[ETH_ALEN];
165 unsigned short eth_proto;
166 };
167
168 struct {
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
175 struct {
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
182 struct {
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
189 struct {
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
196 struct {
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
get_packet_hash(struct packet_description * pckt,bool ipv6)203 static __noinline __u32 get_packet_hash(struct packet_description *pckt, bool ipv6)
204 {
205 if (ipv6)
206 return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
207 pckt->ports, CH_RINGS_SIZE);
208 else
209 return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
210 }
211
get_packet_dst(struct real_definition ** real,struct packet_description * pckt,struct vip_meta * vip_info,bool is_ipv6)212 static __noinline bool get_packet_dst(struct real_definition **real,
213 struct packet_description *pckt,
214 struct vip_meta *vip_info,
215 bool is_ipv6)
216 {
217 __u32 hash = get_packet_hash(pckt, is_ipv6);
218 __u32 key = RING_SIZE * vip_info->vip_num + hash % RING_SIZE;
219 __u32 *real_pos;
220
221 if (hash != 0x358459b7 /* jhash of ipv4 packet */ &&
222 hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
223 return false;
224
225 real_pos = bpf_map_lookup_elem(&ch_rings, &key);
226 if (!real_pos)
227 return false;
228 key = *real_pos;
229 *real = bpf_map_lookup_elem(&reals, &key);
230 if (!(*real))
231 return false;
232 return true;
233 }
234
parse_icmpv6(struct bpf_dynptr * skb_ptr,__u64 off,struct packet_description * pckt)235 static __noinline int parse_icmpv6(struct bpf_dynptr *skb_ptr, __u64 off,
236 struct packet_description *pckt)
237 {
238 __u8 buffer[sizeof(struct ipv6hdr)] = {};
239 struct icmp6hdr *icmp_hdr;
240 struct ipv6hdr *ip6h;
241
242 icmp_hdr = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
243 if (!icmp_hdr)
244 return TC_ACT_SHOT;
245
246 if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
247 return TC_ACT_OK;
248 off += sizeof(struct icmp6hdr);
249 ip6h = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
250 if (!ip6h)
251 return TC_ACT_SHOT;
252 pckt->proto = ip6h->nexthdr;
253 pckt->flags |= F_ICMP;
254 memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
255 memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
256 return TC_ACT_UNSPEC;
257 }
258
parse_icmp(struct bpf_dynptr * skb_ptr,__u64 off,struct packet_description * pckt)259 static __noinline int parse_icmp(struct bpf_dynptr *skb_ptr, __u64 off,
260 struct packet_description *pckt)
261 {
262 __u8 buffer_icmp[sizeof(struct iphdr)] = {};
263 __u8 buffer_ip[sizeof(struct iphdr)] = {};
264 struct icmphdr *icmp_hdr;
265 struct iphdr *iph;
266
267 icmp_hdr = bpf_dynptr_slice(skb_ptr, off, buffer_icmp, sizeof(buffer_icmp));
268 if (!icmp_hdr)
269 return TC_ACT_SHOT;
270 if (icmp_hdr->type != ICMP_DEST_UNREACH ||
271 icmp_hdr->code != ICMP_FRAG_NEEDED)
272 return TC_ACT_OK;
273 off += sizeof(struct icmphdr);
274 iph = bpf_dynptr_slice(skb_ptr, off, buffer_ip, sizeof(buffer_ip));
275 if (!iph || iph->ihl != 5)
276 return TC_ACT_SHOT;
277 pckt->proto = iph->protocol;
278 pckt->flags |= F_ICMP;
279 pckt->src = iph->daddr;
280 pckt->dst = iph->saddr;
281 return TC_ACT_UNSPEC;
282 }
283
parse_udp(struct bpf_dynptr * skb_ptr,__u64 off,struct packet_description * pckt)284 static __noinline bool parse_udp(struct bpf_dynptr *skb_ptr, __u64 off,
285 struct packet_description *pckt)
286 {
287 __u8 buffer[sizeof(struct udphdr)] = {};
288 struct udphdr *udp;
289
290 udp = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
291 if (!udp)
292 return false;
293
294 if (!(pckt->flags & F_ICMP)) {
295 pckt->port16[0] = udp->source;
296 pckt->port16[1] = udp->dest;
297 } else {
298 pckt->port16[0] = udp->dest;
299 pckt->port16[1] = udp->source;
300 }
301 return true;
302 }
303
parse_tcp(struct bpf_dynptr * skb_ptr,__u64 off,struct packet_description * pckt)304 static __noinline bool parse_tcp(struct bpf_dynptr *skb_ptr, __u64 off,
305 struct packet_description *pckt)
306 {
307 __u8 buffer[sizeof(struct tcphdr)] = {};
308 struct tcphdr *tcp;
309
310 tcp = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
311 if (!tcp)
312 return false;
313
314 if (tcp->syn)
315 pckt->flags |= F_SYN_SET;
316
317 if (!(pckt->flags & F_ICMP)) {
318 pckt->port16[0] = tcp->source;
319 pckt->port16[1] = tcp->dest;
320 } else {
321 pckt->port16[0] = tcp->dest;
322 pckt->port16[1] = tcp->source;
323 }
324 return true;
325 }
326
process_packet(struct bpf_dynptr * skb_ptr,struct eth_hdr * eth,__u64 off,bool is_ipv6,struct __sk_buff * skb)327 static __noinline int process_packet(struct bpf_dynptr *skb_ptr,
328 struct eth_hdr *eth, __u64 off,
329 bool is_ipv6, struct __sk_buff *skb)
330 {
331 struct packet_description pckt = {};
332 struct bpf_tunnel_key tkey = {};
333 struct vip_stats *data_stats;
334 struct real_definition *dst;
335 struct vip_meta *vip_info;
336 struct ctl_value *cval;
337 __u32 v4_intf_pos = 1;
338 __u32 v6_intf_pos = 2;
339 struct ipv6hdr *ip6h;
340 struct vip vip = {};
341 struct iphdr *iph;
342 int tun_flag = 0;
343 __u16 pkt_bytes;
344 __u64 iph_len;
345 __u32 ifindex;
346 __u8 protocol;
347 __u32 vip_num;
348 int action;
349
350 tkey.tunnel_ttl = 64;
351 if (is_ipv6) {
352 __u8 buffer[sizeof(struct ipv6hdr)] = {};
353
354 ip6h = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
355 if (!ip6h)
356 return TC_ACT_SHOT;
357
358 iph_len = sizeof(struct ipv6hdr);
359 protocol = ip6h->nexthdr;
360 pckt.proto = protocol;
361 pkt_bytes = bpf_ntohs(ip6h->payload_len);
362 off += iph_len;
363 if (protocol == IPPROTO_FRAGMENT) {
364 return TC_ACT_SHOT;
365 } else if (protocol == IPPROTO_ICMPV6) {
366 action = parse_icmpv6(skb_ptr, off, &pckt);
367 if (action >= 0)
368 return action;
369 off += IPV6_PLUS_ICMP_HDR;
370 } else {
371 memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
372 memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
373 }
374 } else {
375 __u8 buffer[sizeof(struct iphdr)] = {};
376
377 iph = bpf_dynptr_slice(skb_ptr, off, buffer, sizeof(buffer));
378 if (!iph || iph->ihl != 5)
379 return TC_ACT_SHOT;
380
381 protocol = iph->protocol;
382 pckt.proto = protocol;
383 pkt_bytes = bpf_ntohs(iph->tot_len);
384 off += IPV4_HDR_LEN_NO_OPT;
385
386 if (iph->frag_off & PCKT_FRAGMENTED)
387 return TC_ACT_SHOT;
388 if (protocol == IPPROTO_ICMP) {
389 action = parse_icmp(skb_ptr, off, &pckt);
390 if (action >= 0)
391 return action;
392 off += IPV4_PLUS_ICMP_HDR;
393 } else {
394 pckt.src = iph->saddr;
395 pckt.dst = iph->daddr;
396 }
397 }
398 protocol = pckt.proto;
399
400 if (protocol == IPPROTO_TCP) {
401 if (!parse_tcp(skb_ptr, off, &pckt))
402 return TC_ACT_SHOT;
403 } else if (protocol == IPPROTO_UDP) {
404 if (!parse_udp(skb_ptr, off, &pckt))
405 return TC_ACT_SHOT;
406 } else {
407 return TC_ACT_SHOT;
408 }
409
410 if (is_ipv6)
411 memcpy(vip.daddr.v6, pckt.dstv6, 16);
412 else
413 vip.daddr.v4 = pckt.dst;
414
415 vip.dport = pckt.port16[1];
416 vip.protocol = pckt.proto;
417 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
418 if (!vip_info) {
419 vip.dport = 0;
420 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
421 if (!vip_info)
422 return TC_ACT_SHOT;
423 pckt.port16[1] = 0;
424 }
425
426 if (vip_info->flags & F_HASH_NO_SRC_PORT)
427 pckt.port16[0] = 0;
428
429 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
430 return TC_ACT_SHOT;
431
432 if (dst->flags & F_IPV6) {
433 cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
434 if (!cval)
435 return TC_ACT_SHOT;
436 ifindex = cval->ifindex;
437 memcpy(tkey.remote_ipv6, dst->dstv6, 16);
438 tun_flag = BPF_F_TUNINFO_IPV6;
439 } else {
440 cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
441 if (!cval)
442 return TC_ACT_SHOT;
443 ifindex = cval->ifindex;
444 tkey.remote_ipv4 = dst->dst;
445 }
446 vip_num = vip_info->vip_num;
447 data_stats = bpf_map_lookup_elem(&stats, &vip_num);
448 if (!data_stats)
449 return TC_ACT_SHOT;
450 data_stats->pkts++;
451 data_stats->bytes += pkt_bytes;
452 bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
453 *(u32 *)eth->eth_dest = tkey.remote_ipv4;
454 return bpf_redirect(ifindex, 0);
455 }
456
457 SEC("tc")
balancer_ingress(struct __sk_buff * ctx)458 int balancer_ingress(struct __sk_buff *ctx)
459 {
460 __u8 buffer[sizeof(struct eth_hdr)] = {};
461 struct bpf_dynptr ptr;
462 struct eth_hdr *eth;
463 __u32 eth_proto;
464 __u32 nh_off;
465 int err;
466
467 nh_off = sizeof(struct eth_hdr);
468
469 bpf_dynptr_from_skb(ctx, 0, &ptr);
470 eth = bpf_dynptr_slice_rdwr(&ptr, 0, buffer, sizeof(buffer));
471 if (!eth)
472 return TC_ACT_SHOT;
473 eth_proto = eth->eth_proto;
474 if (eth_proto == bpf_htons(ETH_P_IP))
475 err = process_packet(&ptr, eth, nh_off, false, ctx);
476 else if (eth_proto == bpf_htons(ETH_P_IPV6))
477 err = process_packet(&ptr, eth, nh_off, true, ctx);
478 else
479 return TC_ACT_SHOT;
480
481 if (eth == buffer)
482 bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
483
484 return err;
485 }
486
487 char _license[] SEC("license") = "GPL";
488