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 <bpf/bpf_endian.h>
18 
rol32(__u32 word,unsigned int shift)19 static __always_inline __u32 rol32(__u32 word, unsigned int shift)
20 {
21 	return (word << shift) | (word >> ((-shift) & 31));
22 }
23 
24 /* copy paste of jhash from kernel sources to make sure llvm
25  * can compile it into valid sequence of bpf instructions
26  */
27 #define __jhash_mix(a, b, c)			\
28 {						\
29 	a -= c;  a ^= rol32(c, 4);  c += b;	\
30 	b -= a;  b ^= rol32(a, 6);  a += c;	\
31 	c -= b;  c ^= rol32(b, 8);  b += a;	\
32 	a -= c;  a ^= rol32(c, 16); c += b;	\
33 	b -= a;  b ^= rol32(a, 19); a += c;	\
34 	c -= b;  c ^= rol32(b, 4);  b += a;	\
35 }
36 
37 #define __jhash_final(a, b, c)			\
38 {						\
39 	c ^= b; c -= rol32(b, 14);		\
40 	a ^= c; a -= rol32(c, 11);		\
41 	b ^= a; b -= rol32(a, 25);		\
42 	c ^= b; c -= rol32(b, 16);		\
43 	a ^= c; a -= rol32(c, 4);		\
44 	b ^= a; b -= rol32(a, 14);		\
45 	c ^= b; c -= rol32(b, 24);		\
46 }
47 
48 #define JHASH_INITVAL		0xdeadbeef
49 
50 typedef unsigned int u32;
51 
52 static __noinline
jhash(const void * key,u32 length,u32 initval)53 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 
89 __noinline
__jhash_nwords(u32 a,u32 b,u32 c,u32 initval)90 u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
91 {
92 	a += initval;
93 	b += initval;
94 	c += initval;
95 	__jhash_final(a, b, c);
96 	return c;
97 }
98 
99 __noinline
jhash_2words(u32 a,u32 b,u32 initval)100 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 struct flow_key {
106 	union {
107 		__be32 src;
108 		__be32 srcv6[4];
109 	};
110 	union {
111 		__be32 dst;
112 		__be32 dstv6[4];
113 	};
114 	union {
115 		__u32 ports;
116 		__u16 port16[2];
117 	};
118 	__u8 proto;
119 };
120 
121 struct packet_description {
122 	struct flow_key flow;
123 	__u8 flags;
124 };
125 
126 struct ctl_value {
127 	union {
128 		__u64 value;
129 		__u32 ifindex;
130 		__u8 mac[6];
131 	};
132 };
133 
134 struct vip_definition {
135 	union {
136 		__be32 vip;
137 		__be32 vipv6[4];
138 	};
139 	__u16 port;
140 	__u16 family;
141 	__u8 proto;
142 };
143 
144 struct vip_meta {
145 	__u32 flags;
146 	__u32 vip_num;
147 };
148 
149 struct real_pos_lru {
150 	__u32 pos;
151 	__u64 atime;
152 };
153 
154 struct real_definition {
155 	union {
156 		__be32 dst;
157 		__be32 dstv6[4];
158 	};
159 	__u8 flags;
160 };
161 
162 struct lb_stats {
163 	__u64 v2;
164 	__u64 v1;
165 };
166 
167 struct {
168 	__uint(type, BPF_MAP_TYPE_HASH);
169 	__uint(max_entries, 512);
170 	__type(key, struct vip_definition);
171 	__type(value, struct vip_meta);
172 } vip_map SEC(".maps");
173 
174 struct {
175 	__uint(type, BPF_MAP_TYPE_LRU_HASH);
176 	__uint(max_entries, 300);
177 	__uint(map_flags, 1U << 1);
178 	__type(key, struct flow_key);
179 	__type(value, struct real_pos_lru);
180 } lru_cache SEC(".maps");
181 
182 struct {
183 	__uint(type, BPF_MAP_TYPE_ARRAY);
184 	__uint(max_entries, 12 * 655);
185 	__type(key, __u32);
186 	__type(value, __u32);
187 } ch_rings SEC(".maps");
188 
189 struct {
190 	__uint(type, BPF_MAP_TYPE_ARRAY);
191 	__uint(max_entries, 40);
192 	__type(key, __u32);
193 	__type(value, struct real_definition);
194 } reals SEC(".maps");
195 
196 struct {
197 	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
198 	__uint(max_entries, 515);
199 	__type(key, __u32);
200 	__type(value, struct lb_stats);
201 } stats SEC(".maps");
202 
203 struct {
204 	__uint(type, BPF_MAP_TYPE_ARRAY);
205 	__uint(max_entries, 16);
206 	__type(key, __u32);
207 	__type(value, struct ctl_value);
208 } ctl_array SEC(".maps");
209 
210 struct eth_hdr {
211 	unsigned char eth_dest[6];
212 	unsigned char eth_source[6];
213 	unsigned short eth_proto;
214 };
215 
calc_offset(bool is_ipv6,bool is_icmp)216 static __noinline __u64 calc_offset(bool is_ipv6, bool is_icmp)
217 {
218 	__u64 off = sizeof(struct eth_hdr);
219 	if (is_ipv6) {
220 		off += sizeof(struct ipv6hdr);
221 		if (is_icmp)
222 			off += sizeof(struct icmp6hdr) + sizeof(struct ipv6hdr);
223 	} else {
224 		off += sizeof(struct iphdr);
225 		if (is_icmp)
226 			off += sizeof(struct icmphdr) + sizeof(struct iphdr);
227 	}
228 	return off;
229 }
230 
231 static __attribute__ ((noinline))
parse_udp(void * data,void * data_end,bool is_ipv6,struct packet_description * pckt)232 bool parse_udp(void *data, void *data_end,
233 	       bool is_ipv6, struct packet_description *pckt)
234 {
235 
236 	bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
237 	__u64 off = calc_offset(is_ipv6, is_icmp);
238 	struct udphdr *udp;
239 	udp = data + off;
240 
241 	if (udp + 1 > data_end)
242 		return 0;
243 	if (!is_icmp) {
244 		pckt->flow.port16[0] = udp->source;
245 		pckt->flow.port16[1] = udp->dest;
246 	} else {
247 		pckt->flow.port16[0] = udp->dest;
248 		pckt->flow.port16[1] = udp->source;
249 	}
250 	return 1;
251 }
252 
253 static __attribute__ ((noinline))
parse_tcp(void * data,void * data_end,bool is_ipv6,struct packet_description * pckt)254 bool parse_tcp(void *data, void *data_end,
255 	       bool is_ipv6, struct packet_description *pckt)
256 {
257 
258 	bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
259 	__u64 off = calc_offset(is_ipv6, is_icmp);
260 	struct tcphdr *tcp;
261 
262 	tcp = data + off;
263 	if (tcp + 1 > data_end)
264 		return 0;
265 	if (tcp->syn)
266 		pckt->flags |= (1 << 1);
267 	if (!is_icmp) {
268 		pckt->flow.port16[0] = tcp->source;
269 		pckt->flow.port16[1] = tcp->dest;
270 	} else {
271 		pckt->flow.port16[0] = tcp->dest;
272 		pckt->flow.port16[1] = tcp->source;
273 	}
274 	return 1;
275 }
276 
277 static __attribute__ ((noinline))
encap_v6(struct xdp_md * xdp,struct ctl_value * cval,struct packet_description * pckt,struct real_definition * dst,__u32 pkt_bytes)278 bool encap_v6(struct xdp_md *xdp, struct ctl_value *cval,
279 	      struct packet_description *pckt,
280 	      struct real_definition *dst, __u32 pkt_bytes)
281 {
282 	struct eth_hdr *new_eth;
283 	struct eth_hdr *old_eth;
284 	struct ipv6hdr *ip6h;
285 	__u32 ip_suffix;
286 	void *data_end;
287 	void *data;
288 
289 	if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr)))
290 		return 0;
291 	data = (void *)(long)xdp->data;
292 	data_end = (void *)(long)xdp->data_end;
293 	new_eth = data;
294 	ip6h = data + sizeof(struct eth_hdr);
295 	old_eth = data + sizeof(struct ipv6hdr);
296 	if (new_eth + 1 > data_end ||
297 	    old_eth + 1 > data_end || ip6h + 1 > data_end)
298 		return 0;
299 	memcpy(new_eth->eth_dest, cval->mac, 6);
300 	memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
301 	new_eth->eth_proto = 56710;
302 	ip6h->version = 6;
303 	ip6h->priority = 0;
304 	memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
305 
306 	ip6h->nexthdr = IPPROTO_IPV6;
307 	ip_suffix = pckt->flow.srcv6[3] ^ pckt->flow.port16[0];
308 	ip6h->payload_len =
309 	    bpf_htons(pkt_bytes + sizeof(struct ipv6hdr));
310 	ip6h->hop_limit = 4;
311 
312 	ip6h->saddr.in6_u.u6_addr32[0] = 1;
313 	ip6h->saddr.in6_u.u6_addr32[1] = 2;
314 	ip6h->saddr.in6_u.u6_addr32[2] = 3;
315 	ip6h->saddr.in6_u.u6_addr32[3] = ip_suffix;
316 	memcpy(ip6h->daddr.in6_u.u6_addr32, dst->dstv6, 16);
317 	return 1;
318 }
319 
320 static __attribute__ ((noinline))
encap_v4(struct xdp_md * xdp,struct ctl_value * cval,struct packet_description * pckt,struct real_definition * dst,__u32 pkt_bytes)321 bool encap_v4(struct xdp_md *xdp, struct ctl_value *cval,
322 	      struct packet_description *pckt,
323 	      struct real_definition *dst, __u32 pkt_bytes)
324 {
325 
326 	__u32 ip_suffix = bpf_ntohs(pckt->flow.port16[0]);
327 	struct eth_hdr *new_eth;
328 	struct eth_hdr *old_eth;
329 	__u16 *next_iph_u16;
330 	struct iphdr *iph;
331 	__u32 csum = 0;
332 	void *data_end;
333 	void *data;
334 
335 	ip_suffix <<= 15;
336 	ip_suffix ^= pckt->flow.src;
337 	if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr)))
338 		return 0;
339 	data = (void *)(long)xdp->data;
340 	data_end = (void *)(long)xdp->data_end;
341 	new_eth = data;
342 	iph = data + sizeof(struct eth_hdr);
343 	old_eth = data + sizeof(struct iphdr);
344 	if (new_eth + 1 > data_end ||
345 	    old_eth + 1 > data_end || iph + 1 > data_end)
346 		return 0;
347 	memcpy(new_eth->eth_dest, cval->mac, 6);
348 	memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
349 	new_eth->eth_proto = 8;
350 	iph->version = 4;
351 	iph->ihl = 5;
352 	iph->frag_off = 0;
353 	iph->protocol = IPPROTO_IPIP;
354 	iph->check = 0;
355 	iph->tos = 1;
356 	iph->tot_len = bpf_htons(pkt_bytes + sizeof(struct iphdr));
357 	/* don't update iph->daddr, since it will overwrite old eth_proto
358 	 * and multiple iterations of bpf_prog_run() will fail
359 	 */
360 
361 	iph->saddr = ((0xFFFF0000 & ip_suffix) | 4268) ^ dst->dst;
362 	iph->ttl = 4;
363 
364 	next_iph_u16 = (__u16 *) iph;
365 #pragma clang loop unroll(full)
366 	for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
367 		csum += *next_iph_u16++;
368 	iph->check = ~((csum & 0xffff) + (csum >> 16));
369 	if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct iphdr)))
370 		return 0;
371 	return 1;
372 }
373 
374 static __attribute__ ((noinline))
decap_v6(struct xdp_md * xdp,void ** data,void ** data_end,bool inner_v4)375 bool decap_v6(struct xdp_md *xdp, void **data, void **data_end, bool inner_v4)
376 {
377 	struct eth_hdr *new_eth;
378 	struct eth_hdr *old_eth;
379 
380 	old_eth = *data;
381 	new_eth = *data + sizeof(struct ipv6hdr);
382 	memcpy(new_eth->eth_source, old_eth->eth_source, 6);
383 	memcpy(new_eth->eth_dest, old_eth->eth_dest, 6);
384 	if (inner_v4)
385 		new_eth->eth_proto = 8;
386 	else
387 		new_eth->eth_proto = 56710;
388 	if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct ipv6hdr)))
389 		return 0;
390 	*data = (void *)(long)xdp->data;
391 	*data_end = (void *)(long)xdp->data_end;
392 	return 1;
393 }
394 
395 static __attribute__ ((noinline))
decap_v4(struct xdp_md * xdp,void ** data,void ** data_end)396 bool decap_v4(struct xdp_md *xdp, void **data, void **data_end)
397 {
398 	struct eth_hdr *new_eth;
399 	struct eth_hdr *old_eth;
400 
401 	old_eth = *data;
402 	new_eth = *data + sizeof(struct iphdr);
403 	memcpy(new_eth->eth_source, old_eth->eth_source, 6);
404 	memcpy(new_eth->eth_dest, old_eth->eth_dest, 6);
405 	new_eth->eth_proto = 8;
406 	if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct iphdr)))
407 		return 0;
408 	*data = (void *)(long)xdp->data;
409 	*data_end = (void *)(long)xdp->data_end;
410 	return 1;
411 }
412 
413 static __attribute__ ((noinline))
swap_mac_and_send(void * data,void * data_end)414 int swap_mac_and_send(void *data, void *data_end)
415 {
416 	unsigned char tmp_mac[6];
417 	struct eth_hdr *eth;
418 
419 	eth = data;
420 	memcpy(tmp_mac, eth->eth_source, 6);
421 	memcpy(eth->eth_source, eth->eth_dest, 6);
422 	memcpy(eth->eth_dest, tmp_mac, 6);
423 	return XDP_TX;
424 }
425 
426 static __attribute__ ((noinline))
send_icmp_reply(void * data,void * data_end)427 int send_icmp_reply(void *data, void *data_end)
428 {
429 	struct icmphdr *icmp_hdr;
430 	__u16 *next_iph_u16;
431 	__u32 tmp_addr = 0;
432 	struct iphdr *iph;
433 	__u32 csum1 = 0;
434 	__u32 csum = 0;
435 	__u64 off = 0;
436 
437 	if (data + sizeof(struct eth_hdr)
438 	     + sizeof(struct iphdr) + sizeof(struct icmphdr) > data_end)
439 		return XDP_DROP;
440 	off += sizeof(struct eth_hdr);
441 	iph = data + off;
442 	off += sizeof(struct iphdr);
443 	icmp_hdr = data + off;
444 	icmp_hdr->type = 0;
445 	icmp_hdr->checksum += 0x0007;
446 	iph->ttl = 4;
447 	tmp_addr = iph->daddr;
448 	iph->daddr = iph->saddr;
449 	iph->saddr = tmp_addr;
450 	iph->check = 0;
451 	next_iph_u16 = (__u16 *) iph;
452 #pragma clang loop unroll(full)
453 	for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
454 		csum += *next_iph_u16++;
455 	iph->check = ~((csum & 0xffff) + (csum >> 16));
456 	return swap_mac_and_send(data, data_end);
457 }
458 
459 static __attribute__ ((noinline))
send_icmp6_reply(void * data,void * data_end)460 int send_icmp6_reply(void *data, void *data_end)
461 {
462 	struct icmp6hdr *icmp_hdr;
463 	struct ipv6hdr *ip6h;
464 	__be32 tmp_addr[4];
465 	__u64 off = 0;
466 
467 	if (data + sizeof(struct eth_hdr)
468 	     + sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr) > data_end)
469 		return XDP_DROP;
470 	off += sizeof(struct eth_hdr);
471 	ip6h = data + off;
472 	off += sizeof(struct ipv6hdr);
473 	icmp_hdr = data + off;
474 	icmp_hdr->icmp6_type = 129;
475 	icmp_hdr->icmp6_cksum -= 0x0001;
476 	ip6h->hop_limit = 4;
477 	memcpy(tmp_addr, ip6h->saddr.in6_u.u6_addr32, 16);
478 	memcpy(ip6h->saddr.in6_u.u6_addr32, ip6h->daddr.in6_u.u6_addr32, 16);
479 	memcpy(ip6h->daddr.in6_u.u6_addr32, tmp_addr, 16);
480 	return swap_mac_and_send(data, data_end);
481 }
482 
483 static __attribute__ ((noinline))
parse_icmpv6(void * data,void * data_end,__u64 off,struct packet_description * pckt)484 int parse_icmpv6(void *data, void *data_end, __u64 off,
485 		 struct packet_description *pckt)
486 {
487 	struct icmp6hdr *icmp_hdr;
488 	struct ipv6hdr *ip6h;
489 
490 	icmp_hdr = data + off;
491 	if (icmp_hdr + 1 > data_end)
492 		return XDP_DROP;
493 	if (icmp_hdr->icmp6_type == 128)
494 		return send_icmp6_reply(data, data_end);
495 	if (icmp_hdr->icmp6_type != 3)
496 		return XDP_PASS;
497 	off += sizeof(struct icmp6hdr);
498 	ip6h = data + off;
499 	if (ip6h + 1 > data_end)
500 		return XDP_DROP;
501 	pckt->flow.proto = ip6h->nexthdr;
502 	pckt->flags |= (1 << 0);
503 	memcpy(pckt->flow.srcv6, ip6h->daddr.in6_u.u6_addr32, 16);
504 	memcpy(pckt->flow.dstv6, ip6h->saddr.in6_u.u6_addr32, 16);
505 	return -1;
506 }
507 
508 static __attribute__ ((noinline))
parse_icmp(void * data,void * data_end,__u64 off,struct packet_description * pckt)509 int parse_icmp(void *data, void *data_end, __u64 off,
510 	       struct packet_description *pckt)
511 {
512 	struct icmphdr *icmp_hdr;
513 	struct iphdr *iph;
514 
515 	icmp_hdr = data + off;
516 	if (icmp_hdr + 1 > data_end)
517 		return XDP_DROP;
518 	if (icmp_hdr->type == 8)
519 		return send_icmp_reply(data, data_end);
520 	if ((icmp_hdr->type != 3) || (icmp_hdr->code != 4))
521 		return XDP_PASS;
522 	off += sizeof(struct icmphdr);
523 	iph = data + off;
524 	if (iph + 1 > data_end)
525 		return XDP_DROP;
526 	if (iph->ihl != 5)
527 		return XDP_DROP;
528 	pckt->flow.proto = iph->protocol;
529 	pckt->flags |= (1 << 0);
530 	pckt->flow.src = iph->daddr;
531 	pckt->flow.dst = iph->saddr;
532 	return -1;
533 }
534 
535 static __attribute__ ((noinline))
get_packet_hash(struct packet_description * pckt,bool hash_16bytes)536 __u32 get_packet_hash(struct packet_description *pckt,
537 		      bool hash_16bytes)
538 {
539 	if (hash_16bytes)
540 		return jhash_2words(jhash(pckt->flow.srcv6, 16, 12),
541 				    pckt->flow.ports, 24);
542 	else
543 		return jhash_2words(pckt->flow.src, pckt->flow.ports,
544 				    24);
545 }
546 
547 __attribute__ ((noinline))
get_packet_dst(struct real_definition ** real,struct packet_description * pckt,struct vip_meta * vip_info,bool is_ipv6,void * lru_map)548 static bool get_packet_dst(struct real_definition **real,
549 			   struct packet_description *pckt,
550 			   struct vip_meta *vip_info,
551 			   bool is_ipv6, void *lru_map)
552 {
553 	struct real_pos_lru new_dst_lru = { };
554 	bool hash_16bytes = is_ipv6;
555 	__u32 *real_pos, hash, key;
556 	__u64 cur_time;
557 
558 	if (vip_info->flags & (1 << 2))
559 		hash_16bytes = 1;
560 	if (vip_info->flags & (1 << 3)) {
561 		pckt->flow.port16[0] = pckt->flow.port16[1];
562 		memset(pckt->flow.srcv6, 0, 16);
563 	}
564 	hash = get_packet_hash(pckt, hash_16bytes);
565 	if (hash != 0x358459b7 /* jhash of ipv4 packet */  &&
566 	    hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
567 		return false;
568 	key = 2 * vip_info->vip_num + hash % 2;
569 	real_pos = bpf_map_lookup_elem(&ch_rings, &key);
570 	if (!real_pos)
571 		return false;
572 	key = *real_pos;
573 	*real = bpf_map_lookup_elem(&reals, &key);
574 	if (!(*real))
575 		return false;
576 	if (!(vip_info->flags & (1 << 1))) {
577 		__u32 conn_rate_key = 512 + 2;
578 		struct lb_stats *conn_rate_stats =
579 		    bpf_map_lookup_elem(&stats, &conn_rate_key);
580 
581 		if (!conn_rate_stats)
582 			return true;
583 		cur_time = bpf_ktime_get_ns();
584 		if ((cur_time - conn_rate_stats->v2) >> 32 > 0xffFFFF) {
585 			conn_rate_stats->v1 = 1;
586 			conn_rate_stats->v2 = cur_time;
587 		} else {
588 			conn_rate_stats->v1 += 1;
589 			if (conn_rate_stats->v1 >= 1)
590 				return true;
591 		}
592 		if (pckt->flow.proto == IPPROTO_UDP)
593 			new_dst_lru.atime = cur_time;
594 		new_dst_lru.pos = key;
595 		bpf_map_update_elem(lru_map, &pckt->flow, &new_dst_lru, 0);
596 	}
597 	return true;
598 }
599 
600 __attribute__ ((noinline))
connection_table_lookup(struct real_definition ** real,struct packet_description * pckt,void * lru_map)601 static void connection_table_lookup(struct real_definition **real,
602 				    struct packet_description *pckt,
603 				    void *lru_map)
604 {
605 
606 	struct real_pos_lru *dst_lru;
607 	__u64 cur_time;
608 	__u32 key;
609 
610 	dst_lru = bpf_map_lookup_elem(lru_map, &pckt->flow);
611 	if (!dst_lru)
612 		return;
613 	if (pckt->flow.proto == IPPROTO_UDP) {
614 		cur_time = bpf_ktime_get_ns();
615 		if (cur_time - dst_lru->atime > 300000)
616 			return;
617 		dst_lru->atime = cur_time;
618 	}
619 	key = dst_lru->pos;
620 	*real = bpf_map_lookup_elem(&reals, &key);
621 }
622 
623 /* don't believe your eyes!
624  * below function has 6 arguments whereas bpf and llvm allow maximum of 5
625  * but since it's _static_ llvm can optimize one argument away
626  */
627 __attribute__ ((noinline))
process_l3_headers_v6(struct packet_description * pckt,__u8 * protocol,__u64 off,__u16 * pkt_bytes,void * data,void * data_end)628 static int process_l3_headers_v6(struct packet_description *pckt,
629 				 __u8 *protocol, __u64 off,
630 				 __u16 *pkt_bytes, void *data,
631 				 void *data_end)
632 {
633 	struct ipv6hdr *ip6h;
634 	__u64 iph_len;
635 	int action;
636 
637 	ip6h = data + off;
638 	if (ip6h + 1 > data_end)
639 		return XDP_DROP;
640 	iph_len = sizeof(struct ipv6hdr);
641 	*protocol = ip6h->nexthdr;
642 	pckt->flow.proto = *protocol;
643 	*pkt_bytes = bpf_ntohs(ip6h->payload_len);
644 	off += iph_len;
645 	if (*protocol == 45) {
646 		return XDP_DROP;
647 	} else if (*protocol == 59) {
648 		action = parse_icmpv6(data, data_end, off, pckt);
649 		if (action >= 0)
650 			return action;
651 	} else {
652 		memcpy(pckt->flow.srcv6, ip6h->saddr.in6_u.u6_addr32, 16);
653 		memcpy(pckt->flow.dstv6, ip6h->daddr.in6_u.u6_addr32, 16);
654 	}
655 	return -1;
656 }
657 
658 __attribute__ ((noinline))
process_l3_headers_v4(struct packet_description * pckt,__u8 * protocol,__u64 off,__u16 * pkt_bytes,void * data,void * data_end)659 static int process_l3_headers_v4(struct packet_description *pckt,
660 				 __u8 *protocol, __u64 off,
661 				 __u16 *pkt_bytes, void *data,
662 				 void *data_end)
663 {
664 	struct iphdr *iph;
665 	__u64 iph_len;
666 	int action;
667 
668 	iph = data + off;
669 	if (iph + 1 > data_end)
670 		return XDP_DROP;
671 	if (iph->ihl != 5)
672 		return XDP_DROP;
673 	*protocol = iph->protocol;
674 	pckt->flow.proto = *protocol;
675 	*pkt_bytes = bpf_ntohs(iph->tot_len);
676 	off += 20;
677 	if (iph->frag_off & 65343)
678 		return XDP_DROP;
679 	if (*protocol == IPPROTO_ICMP) {
680 		action = parse_icmp(data, data_end, off, pckt);
681 		if (action >= 0)
682 			return action;
683 	} else {
684 		pckt->flow.src = iph->saddr;
685 		pckt->flow.dst = iph->daddr;
686 	}
687 	return -1;
688 }
689 
690 __attribute__ ((noinline))
process_packet(void * data,__u64 off,void * data_end,bool is_ipv6,struct xdp_md * xdp)691 static int process_packet(void *data, __u64 off, void *data_end,
692 			  bool is_ipv6, struct xdp_md *xdp)
693 {
694 
695 	struct real_definition *dst = NULL;
696 	struct packet_description pckt = { };
697 	struct vip_definition vip = { };
698 	struct lb_stats *data_stats;
699 	struct eth_hdr *eth = data;
700 	void *lru_map = &lru_cache;
701 	struct vip_meta *vip_info;
702 	__u32 lru_stats_key = 513;
703 	__u32 mac_addr_pos = 0;
704 	__u32 stats_key = 512;
705 	struct ctl_value *cval;
706 	__u16 pkt_bytes;
707 	__u64 iph_len;
708 	__u8 protocol;
709 	__u32 vip_num;
710 	int action;
711 
712 	if (is_ipv6)
713 		action = process_l3_headers_v6(&pckt, &protocol, off,
714 					       &pkt_bytes, data, data_end);
715 	else
716 		action = process_l3_headers_v4(&pckt, &protocol, off,
717 					       &pkt_bytes, data, data_end);
718 	if (action >= 0)
719 		return action;
720 	protocol = pckt.flow.proto;
721 	if (protocol == IPPROTO_TCP) {
722 		if (!parse_tcp(data, data_end, is_ipv6, &pckt))
723 			return XDP_DROP;
724 	} else if (protocol == IPPROTO_UDP) {
725 		if (!parse_udp(data, data_end, is_ipv6, &pckt))
726 			return XDP_DROP;
727 	} else {
728 		return XDP_TX;
729 	}
730 
731 	if (is_ipv6)
732 		memcpy(vip.vipv6, pckt.flow.dstv6, 16);
733 	else
734 		vip.vip = pckt.flow.dst;
735 	vip.port = pckt.flow.port16[1];
736 	vip.proto = pckt.flow.proto;
737 	vip_info = bpf_map_lookup_elem(&vip_map, &vip);
738 	if (!vip_info) {
739 		vip.port = 0;
740 		vip_info = bpf_map_lookup_elem(&vip_map, &vip);
741 		if (!vip_info)
742 			return XDP_PASS;
743 		if (!(vip_info->flags & (1 << 4)))
744 			pckt.flow.port16[1] = 0;
745 	}
746 	if (data_end - data > 1400)
747 		return XDP_DROP;
748 	data_stats = bpf_map_lookup_elem(&stats, &stats_key);
749 	if (!data_stats)
750 		return XDP_DROP;
751 	data_stats->v1 += 1;
752 	if (!dst) {
753 		if (vip_info->flags & (1 << 0))
754 			pckt.flow.port16[0] = 0;
755 		if (!(pckt.flags & (1 << 1)) && !(vip_info->flags & (1 << 1)))
756 			connection_table_lookup(&dst, &pckt, lru_map);
757 		if (dst)
758 			goto out;
759 		if (pckt.flow.proto == IPPROTO_TCP) {
760 			struct lb_stats *lru_stats =
761 			    bpf_map_lookup_elem(&stats, &lru_stats_key);
762 
763 			if (!lru_stats)
764 				return XDP_DROP;
765 			if (pckt.flags & (1 << 1))
766 				lru_stats->v1 += 1;
767 			else
768 				lru_stats->v2 += 1;
769 		}
770 		if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6, lru_map))
771 			return XDP_DROP;
772 		data_stats->v2 += 1;
773 	}
774 out:
775 	cval = bpf_map_lookup_elem(&ctl_array, &mac_addr_pos);
776 	if (!cval)
777 		return XDP_DROP;
778 	if (dst->flags & (1 << 0)) {
779 		if (!encap_v6(xdp, cval, &pckt, dst, pkt_bytes))
780 			return XDP_DROP;
781 	} else {
782 		if (!encap_v4(xdp, cval, &pckt, dst, pkt_bytes))
783 			return XDP_DROP;
784 	}
785 	vip_num = vip_info->vip_num;
786 	data_stats = bpf_map_lookup_elem(&stats, &vip_num);
787 	if (!data_stats)
788 		return XDP_DROP;
789 	data_stats->v1 += 1;
790 	data_stats->v2 += pkt_bytes;
791 
792 	data = (void *)(long)xdp->data;
793 	data_end = (void *)(long)xdp->data_end;
794 	if (data + 4 > data_end)
795 		return XDP_DROP;
796 	*(u32 *)data = dst->dst;
797 	return XDP_DROP;
798 }
799 
800 SEC("xdp")
balancer_ingress_v4(struct xdp_md * ctx)801 int balancer_ingress_v4(struct xdp_md *ctx)
802 {
803 	void *data = (void *)(long)ctx->data;
804 	void *data_end = (void *)(long)ctx->data_end;
805 	struct eth_hdr *eth = data;
806 	__u32 eth_proto;
807 	__u32 nh_off;
808 
809 	nh_off = sizeof(struct eth_hdr);
810 	if (data + nh_off > data_end)
811 		return XDP_DROP;
812 	eth_proto = bpf_ntohs(eth->eth_proto);
813 	if (eth_proto == ETH_P_IP)
814 		return process_packet(data, nh_off, data_end, 0, ctx);
815 	else
816 		return XDP_DROP;
817 }
818 
819 SEC("xdp")
balancer_ingress_v6(struct xdp_md * ctx)820 int balancer_ingress_v6(struct xdp_md *ctx)
821 {
822 	void *data = (void *)(long)ctx->data;
823 	void *data_end = (void *)(long)ctx->data_end;
824 	struct eth_hdr *eth = data;
825 	__u32 eth_proto;
826 	__u32 nh_off;
827 
828 	nh_off = sizeof(struct eth_hdr);
829 	if (data + nh_off > data_end)
830 		return XDP_DROP;
831 	eth_proto = bpf_ntohs(eth->eth_proto);
832 	if (eth_proto == ETH_P_IPV6)
833 		return process_packet(data, nh_off, data_end, 1, ctx);
834 	else
835 		return XDP_DROP;
836 }
837 
838 char _license[] SEC("license") = "GPL";
839