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 <bpf/bpf_endian.h>
18#include "bpf_compiler.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
54u32 jhash(const void *key, u32 length, u32 initval)
55{
56 u32 a, b, c;
57 const unsigned char *k = key;
58
59 a = b = c = JHASH_INITVAL + length + initval;
60
61 while (length > 12) {
62 a += *(u32 *)(k);
63 b += *(u32 *)(k + 4);
64 c += *(u32 *)(k + 8);
65 __jhash_mix(a, b, c);
66 length -= 12;
67 k += 12;
68 }
69 switch (length) {
70 case 12: c += (u32)k[11]<<24;
71 case 11: c += (u32)k[10]<<16;
72 case 10: c += (u32)k[9]<<8;
73 case 9: c += k[8];
74 case 8: b += (u32)k[7]<<24;
75 case 7: b += (u32)k[6]<<16;
76 case 6: b += (u32)k[5]<<8;
77 case 5: b += k[4];
78 case 4: a += (u32)k[3]<<24;
79 case 3: a += (u32)k[2]<<16;
80 case 2: a += (u32)k[1]<<8;
81 case 1: a += k[0];
82 __jhash_final(a, b, c);
83 case 0: /* Nothing left to add */
84 break;
85 }
86
87 return c;
88}
89
90__noinline
91u32 __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
100__noinline
101u32 jhash_2words(u32 a, u32 b, u32 initval)
102{
103 return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
104}
105
106struct flow_key {
107 union {
108 __be32 src;
109 __be32 srcv6[4];
110 };
111 union {
112 __be32 dst;
113 __be32 dstv6[4];
114 };
115 union {
116 __u32 ports;
117 __u16 port16[2];
118 };
119 __u8 proto;
120};
121
122struct packet_description {
123 struct flow_key flow;
124 __u8 flags;
125};
126
127struct ctl_value {
128 union {
129 __u64 value;
130 __u32 ifindex;
131 __u8 mac[6];
132 };
133};
134
135struct vip_definition {
136 union {
137 __be32 vip;
138 __be32 vipv6[4];
139 };
140 __u16 port;
141 __u16 family;
142 __u8 proto;
143};
144
145struct vip_meta {
146 __u32 flags;
147 __u32 vip_num;
148};
149
150struct real_pos_lru {
151 __u32 pos;
152 __u64 atime;
153};
154
155struct real_definition {
156 union {
157 __be32 dst;
158 __be32 dstv6[4];
159 };
160 __u8 flags;
161};
162
163struct lb_stats {
164 __u64 v2;
165 __u64 v1;
166};
167
168struct {
169 __uint(type, BPF_MAP_TYPE_HASH);
170 __uint(max_entries, 512);
171 __type(key, struct vip_definition);
172 __type(value, struct vip_meta);
173} vip_map SEC(".maps");
174
175struct {
176 __uint(type, BPF_MAP_TYPE_LRU_HASH);
177 __uint(max_entries, 300);
178 __uint(map_flags, 1U << 1);
179 __type(key, struct flow_key);
180 __type(value, struct real_pos_lru);
181} lru_cache SEC(".maps");
182
183struct {
184 __uint(type, BPF_MAP_TYPE_ARRAY);
185 __uint(max_entries, 12 * 655);
186 __type(key, __u32);
187 __type(value, __u32);
188} ch_rings SEC(".maps");
189
190struct {
191 __uint(type, BPF_MAP_TYPE_ARRAY);
192 __uint(max_entries, 40);
193 __type(key, __u32);
194 __type(value, struct real_definition);
195} reals SEC(".maps");
196
197struct {
198 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
199 __uint(max_entries, 515);
200 __type(key, __u32);
201 __type(value, struct lb_stats);
202} stats SEC(".maps");
203
204struct {
205 __uint(type, BPF_MAP_TYPE_ARRAY);
206 __uint(max_entries, 16);
207 __type(key, __u32);
208 __type(value, struct ctl_value);
209} ctl_array SEC(".maps");
210
211struct eth_hdr {
212 unsigned char eth_dest[6];
213 unsigned char eth_source[6];
214 unsigned short eth_proto;
215};
216
217static __noinline __u64 calc_offset(bool is_ipv6, bool is_icmp)
218{
219 __u64 off = sizeof(struct eth_hdr);
220 if (is_ipv6) {
221 off += sizeof(struct ipv6hdr);
222 if (is_icmp)
223 off += sizeof(struct icmp6hdr) + sizeof(struct ipv6hdr);
224 } else {
225 off += sizeof(struct iphdr);
226 if (is_icmp)
227 off += sizeof(struct icmphdr) + sizeof(struct iphdr);
228 }
229 return off;
230}
231
232static __attribute__ ((noinline))
233bool parse_udp(void *data, void *data_end,
234 bool is_ipv6, struct packet_description *pckt)
235{
236
237 bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
238 __u64 off = calc_offset(is_ipv6, is_icmp);
239 struct udphdr *udp;
240 udp = data + off;
241
242 if (udp + 1 > data_end)
243 return false;
244 if (!is_icmp) {
245 pckt->flow.port16[0] = udp->source;
246 pckt->flow.port16[1] = udp->dest;
247 } else {
248 pckt->flow.port16[0] = udp->dest;
249 pckt->flow.port16[1] = udp->source;
250 }
251 return true;
252}
253
254static __attribute__ ((noinline))
255bool parse_tcp(void *data, void *data_end,
256 bool is_ipv6, struct packet_description *pckt)
257{
258
259 bool is_icmp = !((pckt->flags & (1 << 0)) == 0);
260 __u64 off = calc_offset(is_ipv6, is_icmp);
261 struct tcphdr *tcp;
262
263 tcp = data + off;
264 if (tcp + 1 > data_end)
265 return false;
266 if (tcp->syn)
267 pckt->flags |= (1 << 1);
268 if (!is_icmp) {
269 pckt->flow.port16[0] = tcp->source;
270 pckt->flow.port16[1] = tcp->dest;
271 } else {
272 pckt->flow.port16[0] = tcp->dest;
273 pckt->flow.port16[1] = tcp->source;
274 }
275 return true;
276}
277
278static __attribute__ ((noinline))
279bool encap_v6(struct xdp_md *xdp, struct ctl_value *cval,
280 struct packet_description *pckt,
281 struct real_definition *dst, __u32 pkt_bytes)
282{
283 struct eth_hdr *new_eth;
284 struct eth_hdr *old_eth;
285 struct ipv6hdr *ip6h;
286 __u32 ip_suffix;
287 void *data_end;
288 void *data;
289
290 if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr)))
291 return false;
292 data = (void *)(long)xdp->data;
293 data_end = (void *)(long)xdp->data_end;
294 new_eth = data;
295 ip6h = data + sizeof(struct eth_hdr);
296 old_eth = data + sizeof(struct ipv6hdr);
297 if (new_eth + 1 > data_end ||
298 old_eth + 1 > data_end || ip6h + 1 > data_end)
299 return false;
300 memcpy(new_eth->eth_dest, cval->mac, 6);
301 memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
302 new_eth->eth_proto = 56710;
303 ip6h->version = 6;
304 ip6h->priority = 0;
305 memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
306
307 ip6h->nexthdr = IPPROTO_IPV6;
308 ip_suffix = pckt->flow.srcv6[3] ^ pckt->flow.port16[0];
309 ip6h->payload_len =
310 bpf_htons(pkt_bytes + sizeof(struct ipv6hdr));
311 ip6h->hop_limit = 4;
312
313 ip6h->saddr.in6_u.u6_addr32[0] = 1;
314 ip6h->saddr.in6_u.u6_addr32[1] = 2;
315 ip6h->saddr.in6_u.u6_addr32[2] = 3;
316 ip6h->saddr.in6_u.u6_addr32[3] = ip_suffix;
317 memcpy(ip6h->daddr.in6_u.u6_addr32, dst->dstv6, 16);
318 return true;
319}
320
321static __attribute__ ((noinline))
322bool encap_v4(struct xdp_md *xdp, struct ctl_value *cval,
323 struct packet_description *pckt,
324 struct real_definition *dst, __u32 pkt_bytes)
325{
326
327 __u32 ip_suffix = bpf_ntohs(pckt->flow.port16[0]);
328 struct eth_hdr *new_eth;
329 struct eth_hdr *old_eth;
330 __u16 *next_iph_u16;
331 struct iphdr *iph;
332 __u32 csum = 0;
333 void *data_end;
334 void *data;
335
336 ip_suffix <<= 15;
337 ip_suffix ^= pckt->flow.src;
338 if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr)))
339 return false;
340 data = (void *)(long)xdp->data;
341 data_end = (void *)(long)xdp->data_end;
342 new_eth = data;
343 iph = data + sizeof(struct eth_hdr);
344 old_eth = data + sizeof(struct iphdr);
345 if (new_eth + 1 > data_end ||
346 old_eth + 1 > data_end || iph + 1 > data_end)
347 return false;
348 memcpy(new_eth->eth_dest, cval->mac, 6);
349 memcpy(new_eth->eth_source, old_eth->eth_dest, 6);
350 new_eth->eth_proto = 8;
351 iph->version = 4;
352 iph->ihl = 5;
353 iph->frag_off = 0;
354 iph->protocol = IPPROTO_IPIP;
355 iph->check = 0;
356 iph->tos = 1;
357 iph->tot_len = bpf_htons(pkt_bytes + sizeof(struct iphdr));
358 /* don't update iph->daddr, since it will overwrite old eth_proto
359 * and multiple iterations of bpf_prog_run() will fail
360 */
361
362 iph->saddr = ((0xFFFF0000 & ip_suffix) | 4268) ^ dst->dst;
363 iph->ttl = 4;
364
365 next_iph_u16 = (__u16 *) iph;
366 __pragma_loop_unroll_full
367 for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
368 csum += *next_iph_u16++;
369 iph->check = ~((csum & 0xffff) + (csum >> 16));
370 if (bpf_xdp_adjust_head(xdp, (int)sizeof(struct iphdr)))
371 return false;
372 return true;
373}
374
375static __attribute__ ((noinline))
376int swap_mac_and_send(void *data, void *data_end)
377{
378 unsigned char tmp_mac[6];
379 struct eth_hdr *eth;
380
381 eth = data;
382 memcpy(tmp_mac, eth->eth_source, 6);
383 memcpy(eth->eth_source, eth->eth_dest, 6);
384 memcpy(eth->eth_dest, tmp_mac, 6);
385 return XDP_TX;
386}
387
388static __attribute__ ((noinline))
389int send_icmp_reply(void *data, void *data_end)
390{
391 struct icmphdr *icmp_hdr;
392 __u16 *next_iph_u16;
393 __u32 tmp_addr = 0;
394 struct iphdr *iph;
395 __u32 csum = 0;
396 __u64 off = 0;
397
398 if (data + sizeof(struct eth_hdr)
399 + sizeof(struct iphdr) + sizeof(struct icmphdr) > data_end)
400 return XDP_DROP;
401 off += sizeof(struct eth_hdr);
402 iph = data + off;
403 off += sizeof(struct iphdr);
404 icmp_hdr = data + off;
405 icmp_hdr->type = 0;
406 icmp_hdr->checksum += 0x0007;
407 iph->ttl = 4;
408 tmp_addr = iph->daddr;
409 iph->daddr = iph->saddr;
410 iph->saddr = tmp_addr;
411 iph->check = 0;
412 next_iph_u16 = (__u16 *) iph;
413 __pragma_loop_unroll_full
414 for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
415 csum += *next_iph_u16++;
416 iph->check = ~((csum & 0xffff) + (csum >> 16));
417 return swap_mac_and_send(data, data_end);
418}
419
420static __attribute__ ((noinline))
421int send_icmp6_reply(void *data, void *data_end)
422{
423 struct icmp6hdr *icmp_hdr;
424 struct ipv6hdr *ip6h;
425 __be32 tmp_addr[4];
426 __u64 off = 0;
427
428 if (data + sizeof(struct eth_hdr)
429 + sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr) > data_end)
430 return XDP_DROP;
431 off += sizeof(struct eth_hdr);
432 ip6h = data + off;
433 off += sizeof(struct ipv6hdr);
434 icmp_hdr = data + off;
435 icmp_hdr->icmp6_type = 129;
436 icmp_hdr->icmp6_cksum -= 0x0001;
437 ip6h->hop_limit = 4;
438 memcpy(tmp_addr, ip6h->saddr.in6_u.u6_addr32, 16);
439 memcpy(ip6h->saddr.in6_u.u6_addr32, ip6h->daddr.in6_u.u6_addr32, 16);
440 memcpy(ip6h->daddr.in6_u.u6_addr32, tmp_addr, 16);
441 return swap_mac_and_send(data, data_end);
442}
443
444static __attribute__ ((noinline))
445int parse_icmpv6(void *data, void *data_end, __u64 off,
446 struct packet_description *pckt)
447{
448 struct icmp6hdr *icmp_hdr;
449 struct ipv6hdr *ip6h;
450
451 icmp_hdr = data + off;
452 if (icmp_hdr + 1 > data_end)
453 return XDP_DROP;
454 if (icmp_hdr->icmp6_type == 128)
455 return send_icmp6_reply(data, data_end);
456 if (icmp_hdr->icmp6_type != 3)
457 return XDP_PASS;
458 off += sizeof(struct icmp6hdr);
459 ip6h = data + off;
460 if (ip6h + 1 > data_end)
461 return XDP_DROP;
462 pckt->flow.proto = ip6h->nexthdr;
463 pckt->flags |= (1 << 0);
464 memcpy(pckt->flow.srcv6, ip6h->daddr.in6_u.u6_addr32, 16);
465 memcpy(pckt->flow.dstv6, ip6h->saddr.in6_u.u6_addr32, 16);
466 return -1;
467}
468
469static __attribute__ ((noinline))
470int parse_icmp(void *data, void *data_end, __u64 off,
471 struct packet_description *pckt)
472{
473 struct icmphdr *icmp_hdr;
474 struct iphdr *iph;
475
476 icmp_hdr = data + off;
477 if (icmp_hdr + 1 > data_end)
478 return XDP_DROP;
479 if (icmp_hdr->type == 8)
480 return send_icmp_reply(data, data_end);
481 if ((icmp_hdr->type != 3) || (icmp_hdr->code != 4))
482 return XDP_PASS;
483 off += sizeof(struct icmphdr);
484 iph = data + off;
485 if (iph + 1 > data_end)
486 return XDP_DROP;
487 if (iph->ihl != 5)
488 return XDP_DROP;
489 pckt->flow.proto = iph->protocol;
490 pckt->flags |= (1 << 0);
491 pckt->flow.src = iph->daddr;
492 pckt->flow.dst = iph->saddr;
493 return -1;
494}
495
496static __attribute__ ((noinline))
497__u32 get_packet_hash(struct packet_description *pckt,
498 bool hash_16bytes)
499{
500 if (hash_16bytes)
501 return jhash_2words(jhash(pckt->flow.srcv6, 16, 12),
502 pckt->flow.ports, 24);
503 else
504 return jhash_2words(pckt->flow.src, pckt->flow.ports,
505 24);
506}
507
508__attribute__ ((noinline))
509static bool get_packet_dst(struct real_definition **real,
510 struct packet_description *pckt,
511 struct vip_meta *vip_info,
512 bool is_ipv6, void *lru_map)
513{
514 struct real_pos_lru new_dst_lru = { };
515 bool hash_16bytes = is_ipv6;
516 __u32 *real_pos, hash, key;
517 __u64 cur_time;
518
519 if (vip_info->flags & (1 << 2))
520 hash_16bytes = 1;
521 if (vip_info->flags & (1 << 3)) {
522 pckt->flow.port16[0] = pckt->flow.port16[1];
523 memset(pckt->flow.srcv6, 0, 16);
524 }
525 hash = get_packet_hash(pckt, hash_16bytes);
526 if (hash != 0x358459b7 /* jhash of ipv4 packet */ &&
527 hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
528 return false;
529 key = 2 * vip_info->vip_num + hash % 2;
530 real_pos = bpf_map_lookup_elem(&ch_rings, &key);
531 if (!real_pos)
532 return false;
533 key = *real_pos;
534 *real = bpf_map_lookup_elem(&reals, &key);
535 if (!(*real))
536 return false;
537 if (!(vip_info->flags & (1 << 1))) {
538 __u32 conn_rate_key = 512 + 2;
539 struct lb_stats *conn_rate_stats =
540 bpf_map_lookup_elem(&stats, &conn_rate_key);
541
542 if (!conn_rate_stats)
543 return true;
544 cur_time = bpf_ktime_get_ns();
545 if ((cur_time - conn_rate_stats->v2) >> 32 > 0xffFFFF) {
546 conn_rate_stats->v1 = 1;
547 conn_rate_stats->v2 = cur_time;
548 } else {
549 conn_rate_stats->v1 += 1;
550 if (conn_rate_stats->v1 >= 1)
551 return true;
552 }
553 if (pckt->flow.proto == IPPROTO_UDP)
554 new_dst_lru.atime = cur_time;
555 new_dst_lru.pos = key;
556 bpf_map_update_elem(lru_map, &pckt->flow, &new_dst_lru, 0);
557 }
558 return true;
559}
560
561__attribute__ ((noinline))
562static void connection_table_lookup(struct real_definition **real,
563 struct packet_description *pckt,
564 void *lru_map)
565{
566
567 struct real_pos_lru *dst_lru;
568 __u64 cur_time;
569 __u32 key;
570
571 dst_lru = bpf_map_lookup_elem(lru_map, &pckt->flow);
572 if (!dst_lru)
573 return;
574 if (pckt->flow.proto == IPPROTO_UDP) {
575 cur_time = bpf_ktime_get_ns();
576 if (cur_time - dst_lru->atime > 300000)
577 return;
578 dst_lru->atime = cur_time;
579 }
580 key = dst_lru->pos;
581 *real = bpf_map_lookup_elem(&reals, &key);
582}
583
584/* don't believe your eyes!
585 * below function has 6 arguments whereas bpf and llvm allow maximum of 5
586 * but since it's _static_ llvm can optimize one argument away
587 */
588__attribute__ ((noinline))
589static int process_l3_headers_v6(struct packet_description *pckt,
590 __u8 *protocol, __u64 off,
591 __u16 *pkt_bytes, void *data,
592 void *data_end)
593{
594 struct ipv6hdr *ip6h;
595 __u64 iph_len;
596 int action;
597
598 ip6h = data + off;
599 if (ip6h + 1 > data_end)
600 return XDP_DROP;
601 iph_len = sizeof(struct ipv6hdr);
602 *protocol = ip6h->nexthdr;
603 pckt->flow.proto = *protocol;
604 *pkt_bytes = bpf_ntohs(ip6h->payload_len);
605 off += iph_len;
606 if (*protocol == 45) {
607 return XDP_DROP;
608 } else if (*protocol == 59) {
609 action = parse_icmpv6(data, data_end, off, pckt);
610 if (action >= 0)
611 return action;
612 } else {
613 memcpy(pckt->flow.srcv6, ip6h->saddr.in6_u.u6_addr32, 16);
614 memcpy(pckt->flow.dstv6, ip6h->daddr.in6_u.u6_addr32, 16);
615 }
616 return -1;
617}
618
619__attribute__ ((noinline))
620static int process_l3_headers_v4(struct packet_description *pckt,
621 __u8 *protocol, __u64 off,
622 __u16 *pkt_bytes, void *data,
623 void *data_end)
624{
625 struct iphdr *iph;
626 int action;
627
628 iph = data + off;
629 if (iph + 1 > data_end)
630 return XDP_DROP;
631 if (iph->ihl != 5)
632 return XDP_DROP;
633 *protocol = iph->protocol;
634 pckt->flow.proto = *protocol;
635 *pkt_bytes = bpf_ntohs(iph->tot_len);
636 off += 20;
637 if (iph->frag_off & 65343)
638 return XDP_DROP;
639 if (*protocol == IPPROTO_ICMP) {
640 action = parse_icmp(data, data_end, off, pckt);
641 if (action >= 0)
642 return action;
643 } else {
644 pckt->flow.src = iph->saddr;
645 pckt->flow.dst = iph->daddr;
646 }
647 return -1;
648}
649
650__attribute__ ((noinline))
651static int process_packet(void *data, __u64 off, void *data_end,
652 bool is_ipv6, struct xdp_md *xdp)
653{
654
655 struct real_definition *dst = NULL;
656 struct packet_description pckt = { };
657 struct vip_definition vip = { };
658 struct lb_stats *data_stats;
659 void *lru_map = &lru_cache;
660 struct vip_meta *vip_info;
661 __u32 lru_stats_key = 513;
662 __u32 mac_addr_pos = 0;
663 __u32 stats_key = 512;
664 struct ctl_value *cval;
665 __u16 pkt_bytes;
666 __u8 protocol;
667 __u32 vip_num;
668 int action;
669
670 if (is_ipv6)
671 action = process_l3_headers_v6(&pckt, &protocol, off,
672 &pkt_bytes, data, data_end);
673 else
674 action = process_l3_headers_v4(&pckt, &protocol, off,
675 &pkt_bytes, data, data_end);
676 if (action >= 0)
677 return action;
678 protocol = pckt.flow.proto;
679 if (protocol == IPPROTO_TCP) {
680 if (!parse_tcp(data, data_end, is_ipv6, &pckt))
681 return XDP_DROP;
682 } else if (protocol == IPPROTO_UDP) {
683 if (!parse_udp(data, data_end, is_ipv6, &pckt))
684 return XDP_DROP;
685 } else {
686 return XDP_TX;
687 }
688
689 if (is_ipv6)
690 memcpy(vip.vipv6, pckt.flow.dstv6, 16);
691 else
692 vip.vip = pckt.flow.dst;
693 vip.port = pckt.flow.port16[1];
694 vip.proto = pckt.flow.proto;
695 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
696 if (!vip_info) {
697 vip.port = 0;
698 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
699 if (!vip_info)
700 return XDP_PASS;
701 if (!(vip_info->flags & (1 << 4)))
702 pckt.flow.port16[1] = 0;
703 }
704 if (data_end - data > 1400)
705 return XDP_DROP;
706 data_stats = bpf_map_lookup_elem(&stats, &stats_key);
707 if (!data_stats)
708 return XDP_DROP;
709 data_stats->v1 += 1;
710 if (!dst) {
711 if (vip_info->flags & (1 << 0))
712 pckt.flow.port16[0] = 0;
713 if (!(pckt.flags & (1 << 1)) && !(vip_info->flags & (1 << 1)))
714 connection_table_lookup(&dst, &pckt, lru_map);
715 if (dst)
716 goto out;
717 if (pckt.flow.proto == IPPROTO_TCP) {
718 struct lb_stats *lru_stats =
719 bpf_map_lookup_elem(&stats, &lru_stats_key);
720
721 if (!lru_stats)
722 return XDP_DROP;
723 if (pckt.flags & (1 << 1))
724 lru_stats->v1 += 1;
725 else
726 lru_stats->v2 += 1;
727 }
728 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6, lru_map))
729 return XDP_DROP;
730 data_stats->v2 += 1;
731 }
732out:
733 cval = bpf_map_lookup_elem(&ctl_array, &mac_addr_pos);
734 if (!cval)
735 return XDP_DROP;
736 if (dst->flags & (1 << 0)) {
737 if (!encap_v6(xdp, cval, &pckt, dst, pkt_bytes))
738 return XDP_DROP;
739 } else {
740 if (!encap_v4(xdp, cval, &pckt, dst, pkt_bytes))
741 return XDP_DROP;
742 }
743 vip_num = vip_info->vip_num;
744 data_stats = bpf_map_lookup_elem(&stats, &vip_num);
745 if (!data_stats)
746 return XDP_DROP;
747 data_stats->v1 += 1;
748 data_stats->v2 += pkt_bytes;
749
750 data = (void *)(long)xdp->data;
751 data_end = (void *)(long)xdp->data_end;
752 if (data + 4 > data_end)
753 return XDP_DROP;
754 *(u32 *)data = dst->dst;
755 return XDP_DROP;
756}
757
758SEC("xdp")
759int balancer_ingress_v4(struct xdp_md *ctx)
760{
761 void *data = (void *)(long)ctx->data;
762 void *data_end = (void *)(long)ctx->data_end;
763 struct eth_hdr *eth = data;
764 __u32 eth_proto;
765 __u32 nh_off;
766
767 nh_off = sizeof(struct eth_hdr);
768 if (data + nh_off > data_end)
769 return XDP_DROP;
770 eth_proto = bpf_ntohs(eth->eth_proto);
771 if (eth_proto == ETH_P_IP)
772 return process_packet(data, nh_off, data_end, 0, ctx);
773 else
774 return XDP_DROP;
775}
776
777SEC("xdp")
778int balancer_ingress_v6(struct xdp_md *ctx)
779{
780 void *data = (void *)(long)ctx->data;
781 void *data_end = (void *)(long)ctx->data_end;
782 struct eth_hdr *eth = data;
783 __u32 eth_proto;
784 __u32 nh_off;
785
786 nh_off = sizeof(struct eth_hdr);
787 if (data + nh_off > data_end)
788 return XDP_DROP;
789 eth_proto = bpf_ntohs(eth->eth_proto);
790 if (eth_proto == ETH_P_IPV6)
791 return process_packet(data, nh_off, data_end, 1, ctx);
792 else
793 return XDP_DROP;
794}
795
796char _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 <bpf/bpf_endian.h>
18
19static __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
50typedef unsigned int u32;
51
52static __noinline
53u32 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
90u32 __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
100u32 jhash_2words(u32 a, u32 b, u32 initval)
101{
102 return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
103}
104
105struct 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
121struct packet_description {
122 struct flow_key flow;
123 __u8 flags;
124};
125
126struct ctl_value {
127 union {
128 __u64 value;
129 __u32 ifindex;
130 __u8 mac[6];
131 };
132};
133
134struct vip_definition {
135 union {
136 __be32 vip;
137 __be32 vipv6[4];
138 };
139 __u16 port;
140 __u16 family;
141 __u8 proto;
142};
143
144struct vip_meta {
145 __u32 flags;
146 __u32 vip_num;
147};
148
149struct real_pos_lru {
150 __u32 pos;
151 __u64 atime;
152};
153
154struct real_definition {
155 union {
156 __be32 dst;
157 __be32 dstv6[4];
158 };
159 __u8 flags;
160};
161
162struct lb_stats {
163 __u64 v2;
164 __u64 v1;
165};
166
167struct {
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
174struct {
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
182struct {
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
189struct {
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
196struct {
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
203struct {
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
210struct eth_hdr {
211 unsigned char eth_dest[6];
212 unsigned char eth_source[6];
213 unsigned short eth_proto;
214};
215
216static __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
231static __attribute__ ((noinline))
232bool 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 false;
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 true;
251}
252
253static __attribute__ ((noinline))
254bool 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 false;
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 true;
275}
276
277static __attribute__ ((noinline))
278bool 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 false;
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 false;
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 true;
318}
319
320static __attribute__ ((noinline))
321bool 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 false;
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 false;
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 false;
371 return true;
372}
373
374static __attribute__ ((noinline))
375int swap_mac_and_send(void *data, void *data_end)
376{
377 unsigned char tmp_mac[6];
378 struct eth_hdr *eth;
379
380 eth = data;
381 memcpy(tmp_mac, eth->eth_source, 6);
382 memcpy(eth->eth_source, eth->eth_dest, 6);
383 memcpy(eth->eth_dest, tmp_mac, 6);
384 return XDP_TX;
385}
386
387static __attribute__ ((noinline))
388int send_icmp_reply(void *data, void *data_end)
389{
390 struct icmphdr *icmp_hdr;
391 __u16 *next_iph_u16;
392 __u32 tmp_addr = 0;
393 struct iphdr *iph;
394 __u32 csum = 0;
395 __u64 off = 0;
396
397 if (data + sizeof(struct eth_hdr)
398 + sizeof(struct iphdr) + sizeof(struct icmphdr) > data_end)
399 return XDP_DROP;
400 off += sizeof(struct eth_hdr);
401 iph = data + off;
402 off += sizeof(struct iphdr);
403 icmp_hdr = data + off;
404 icmp_hdr->type = 0;
405 icmp_hdr->checksum += 0x0007;
406 iph->ttl = 4;
407 tmp_addr = iph->daddr;
408 iph->daddr = iph->saddr;
409 iph->saddr = tmp_addr;
410 iph->check = 0;
411 next_iph_u16 = (__u16 *) iph;
412#pragma clang loop unroll(full)
413 for (int i = 0; i < sizeof(struct iphdr) >> 1; i++)
414 csum += *next_iph_u16++;
415 iph->check = ~((csum & 0xffff) + (csum >> 16));
416 return swap_mac_and_send(data, data_end);
417}
418
419static __attribute__ ((noinline))
420int send_icmp6_reply(void *data, void *data_end)
421{
422 struct icmp6hdr *icmp_hdr;
423 struct ipv6hdr *ip6h;
424 __be32 tmp_addr[4];
425 __u64 off = 0;
426
427 if (data + sizeof(struct eth_hdr)
428 + sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr) > data_end)
429 return XDP_DROP;
430 off += sizeof(struct eth_hdr);
431 ip6h = data + off;
432 off += sizeof(struct ipv6hdr);
433 icmp_hdr = data + off;
434 icmp_hdr->icmp6_type = 129;
435 icmp_hdr->icmp6_cksum -= 0x0001;
436 ip6h->hop_limit = 4;
437 memcpy(tmp_addr, ip6h->saddr.in6_u.u6_addr32, 16);
438 memcpy(ip6h->saddr.in6_u.u6_addr32, ip6h->daddr.in6_u.u6_addr32, 16);
439 memcpy(ip6h->daddr.in6_u.u6_addr32, tmp_addr, 16);
440 return swap_mac_and_send(data, data_end);
441}
442
443static __attribute__ ((noinline))
444int parse_icmpv6(void *data, void *data_end, __u64 off,
445 struct packet_description *pckt)
446{
447 struct icmp6hdr *icmp_hdr;
448 struct ipv6hdr *ip6h;
449
450 icmp_hdr = data + off;
451 if (icmp_hdr + 1 > data_end)
452 return XDP_DROP;
453 if (icmp_hdr->icmp6_type == 128)
454 return send_icmp6_reply(data, data_end);
455 if (icmp_hdr->icmp6_type != 3)
456 return XDP_PASS;
457 off += sizeof(struct icmp6hdr);
458 ip6h = data + off;
459 if (ip6h + 1 > data_end)
460 return XDP_DROP;
461 pckt->flow.proto = ip6h->nexthdr;
462 pckt->flags |= (1 << 0);
463 memcpy(pckt->flow.srcv6, ip6h->daddr.in6_u.u6_addr32, 16);
464 memcpy(pckt->flow.dstv6, ip6h->saddr.in6_u.u6_addr32, 16);
465 return -1;
466}
467
468static __attribute__ ((noinline))
469int parse_icmp(void *data, void *data_end, __u64 off,
470 struct packet_description *pckt)
471{
472 struct icmphdr *icmp_hdr;
473 struct iphdr *iph;
474
475 icmp_hdr = data + off;
476 if (icmp_hdr + 1 > data_end)
477 return XDP_DROP;
478 if (icmp_hdr->type == 8)
479 return send_icmp_reply(data, data_end);
480 if ((icmp_hdr->type != 3) || (icmp_hdr->code != 4))
481 return XDP_PASS;
482 off += sizeof(struct icmphdr);
483 iph = data + off;
484 if (iph + 1 > data_end)
485 return XDP_DROP;
486 if (iph->ihl != 5)
487 return XDP_DROP;
488 pckt->flow.proto = iph->protocol;
489 pckt->flags |= (1 << 0);
490 pckt->flow.src = iph->daddr;
491 pckt->flow.dst = iph->saddr;
492 return -1;
493}
494
495static __attribute__ ((noinline))
496__u32 get_packet_hash(struct packet_description *pckt,
497 bool hash_16bytes)
498{
499 if (hash_16bytes)
500 return jhash_2words(jhash(pckt->flow.srcv6, 16, 12),
501 pckt->flow.ports, 24);
502 else
503 return jhash_2words(pckt->flow.src, pckt->flow.ports,
504 24);
505}
506
507__attribute__ ((noinline))
508static bool get_packet_dst(struct real_definition **real,
509 struct packet_description *pckt,
510 struct vip_meta *vip_info,
511 bool is_ipv6, void *lru_map)
512{
513 struct real_pos_lru new_dst_lru = { };
514 bool hash_16bytes = is_ipv6;
515 __u32 *real_pos, hash, key;
516 __u64 cur_time;
517
518 if (vip_info->flags & (1 << 2))
519 hash_16bytes = 1;
520 if (vip_info->flags & (1 << 3)) {
521 pckt->flow.port16[0] = pckt->flow.port16[1];
522 memset(pckt->flow.srcv6, 0, 16);
523 }
524 hash = get_packet_hash(pckt, hash_16bytes);
525 if (hash != 0x358459b7 /* jhash of ipv4 packet */ &&
526 hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
527 return false;
528 key = 2 * vip_info->vip_num + hash % 2;
529 real_pos = bpf_map_lookup_elem(&ch_rings, &key);
530 if (!real_pos)
531 return false;
532 key = *real_pos;
533 *real = bpf_map_lookup_elem(&reals, &key);
534 if (!(*real))
535 return false;
536 if (!(vip_info->flags & (1 << 1))) {
537 __u32 conn_rate_key = 512 + 2;
538 struct lb_stats *conn_rate_stats =
539 bpf_map_lookup_elem(&stats, &conn_rate_key);
540
541 if (!conn_rate_stats)
542 return true;
543 cur_time = bpf_ktime_get_ns();
544 if ((cur_time - conn_rate_stats->v2) >> 32 > 0xffFFFF) {
545 conn_rate_stats->v1 = 1;
546 conn_rate_stats->v2 = cur_time;
547 } else {
548 conn_rate_stats->v1 += 1;
549 if (conn_rate_stats->v1 >= 1)
550 return true;
551 }
552 if (pckt->flow.proto == IPPROTO_UDP)
553 new_dst_lru.atime = cur_time;
554 new_dst_lru.pos = key;
555 bpf_map_update_elem(lru_map, &pckt->flow, &new_dst_lru, 0);
556 }
557 return true;
558}
559
560__attribute__ ((noinline))
561static void connection_table_lookup(struct real_definition **real,
562 struct packet_description *pckt,
563 void *lru_map)
564{
565
566 struct real_pos_lru *dst_lru;
567 __u64 cur_time;
568 __u32 key;
569
570 dst_lru = bpf_map_lookup_elem(lru_map, &pckt->flow);
571 if (!dst_lru)
572 return;
573 if (pckt->flow.proto == IPPROTO_UDP) {
574 cur_time = bpf_ktime_get_ns();
575 if (cur_time - dst_lru->atime > 300000)
576 return;
577 dst_lru->atime = cur_time;
578 }
579 key = dst_lru->pos;
580 *real = bpf_map_lookup_elem(&reals, &key);
581}
582
583/* don't believe your eyes!
584 * below function has 6 arguments whereas bpf and llvm allow maximum of 5
585 * but since it's _static_ llvm can optimize one argument away
586 */
587__attribute__ ((noinline))
588static int process_l3_headers_v6(struct packet_description *pckt,
589 __u8 *protocol, __u64 off,
590 __u16 *pkt_bytes, void *data,
591 void *data_end)
592{
593 struct ipv6hdr *ip6h;
594 __u64 iph_len;
595 int action;
596
597 ip6h = data + off;
598 if (ip6h + 1 > data_end)
599 return XDP_DROP;
600 iph_len = sizeof(struct ipv6hdr);
601 *protocol = ip6h->nexthdr;
602 pckt->flow.proto = *protocol;
603 *pkt_bytes = bpf_ntohs(ip6h->payload_len);
604 off += iph_len;
605 if (*protocol == 45) {
606 return XDP_DROP;
607 } else if (*protocol == 59) {
608 action = parse_icmpv6(data, data_end, off, pckt);
609 if (action >= 0)
610 return action;
611 } else {
612 memcpy(pckt->flow.srcv6, ip6h->saddr.in6_u.u6_addr32, 16);
613 memcpy(pckt->flow.dstv6, ip6h->daddr.in6_u.u6_addr32, 16);
614 }
615 return -1;
616}
617
618__attribute__ ((noinline))
619static int process_l3_headers_v4(struct packet_description *pckt,
620 __u8 *protocol, __u64 off,
621 __u16 *pkt_bytes, void *data,
622 void *data_end)
623{
624 struct iphdr *iph;
625 int action;
626
627 iph = data + off;
628 if (iph + 1 > data_end)
629 return XDP_DROP;
630 if (iph->ihl != 5)
631 return XDP_DROP;
632 *protocol = iph->protocol;
633 pckt->flow.proto = *protocol;
634 *pkt_bytes = bpf_ntohs(iph->tot_len);
635 off += 20;
636 if (iph->frag_off & 65343)
637 return XDP_DROP;
638 if (*protocol == IPPROTO_ICMP) {
639 action = parse_icmp(data, data_end, off, pckt);
640 if (action >= 0)
641 return action;
642 } else {
643 pckt->flow.src = iph->saddr;
644 pckt->flow.dst = iph->daddr;
645 }
646 return -1;
647}
648
649__attribute__ ((noinline))
650static int process_packet(void *data, __u64 off, void *data_end,
651 bool is_ipv6, struct xdp_md *xdp)
652{
653
654 struct real_definition *dst = NULL;
655 struct packet_description pckt = { };
656 struct vip_definition vip = { };
657 struct lb_stats *data_stats;
658 void *lru_map = &lru_cache;
659 struct vip_meta *vip_info;
660 __u32 lru_stats_key = 513;
661 __u32 mac_addr_pos = 0;
662 __u32 stats_key = 512;
663 struct ctl_value *cval;
664 __u16 pkt_bytes;
665 __u8 protocol;
666 __u32 vip_num;
667 int action;
668
669 if (is_ipv6)
670 action = process_l3_headers_v6(&pckt, &protocol, off,
671 &pkt_bytes, data, data_end);
672 else
673 action = process_l3_headers_v4(&pckt, &protocol, off,
674 &pkt_bytes, data, data_end);
675 if (action >= 0)
676 return action;
677 protocol = pckt.flow.proto;
678 if (protocol == IPPROTO_TCP) {
679 if (!parse_tcp(data, data_end, is_ipv6, &pckt))
680 return XDP_DROP;
681 } else if (protocol == IPPROTO_UDP) {
682 if (!parse_udp(data, data_end, is_ipv6, &pckt))
683 return XDP_DROP;
684 } else {
685 return XDP_TX;
686 }
687
688 if (is_ipv6)
689 memcpy(vip.vipv6, pckt.flow.dstv6, 16);
690 else
691 vip.vip = pckt.flow.dst;
692 vip.port = pckt.flow.port16[1];
693 vip.proto = pckt.flow.proto;
694 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
695 if (!vip_info) {
696 vip.port = 0;
697 vip_info = bpf_map_lookup_elem(&vip_map, &vip);
698 if (!vip_info)
699 return XDP_PASS;
700 if (!(vip_info->flags & (1 << 4)))
701 pckt.flow.port16[1] = 0;
702 }
703 if (data_end - data > 1400)
704 return XDP_DROP;
705 data_stats = bpf_map_lookup_elem(&stats, &stats_key);
706 if (!data_stats)
707 return XDP_DROP;
708 data_stats->v1 += 1;
709 if (!dst) {
710 if (vip_info->flags & (1 << 0))
711 pckt.flow.port16[0] = 0;
712 if (!(pckt.flags & (1 << 1)) && !(vip_info->flags & (1 << 1)))
713 connection_table_lookup(&dst, &pckt, lru_map);
714 if (dst)
715 goto out;
716 if (pckt.flow.proto == IPPROTO_TCP) {
717 struct lb_stats *lru_stats =
718 bpf_map_lookup_elem(&stats, &lru_stats_key);
719
720 if (!lru_stats)
721 return XDP_DROP;
722 if (pckt.flags & (1 << 1))
723 lru_stats->v1 += 1;
724 else
725 lru_stats->v2 += 1;
726 }
727 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6, lru_map))
728 return XDP_DROP;
729 data_stats->v2 += 1;
730 }
731out:
732 cval = bpf_map_lookup_elem(&ctl_array, &mac_addr_pos);
733 if (!cval)
734 return XDP_DROP;
735 if (dst->flags & (1 << 0)) {
736 if (!encap_v6(xdp, cval, &pckt, dst, pkt_bytes))
737 return XDP_DROP;
738 } else {
739 if (!encap_v4(xdp, cval, &pckt, dst, pkt_bytes))
740 return XDP_DROP;
741 }
742 vip_num = vip_info->vip_num;
743 data_stats = bpf_map_lookup_elem(&stats, &vip_num);
744 if (!data_stats)
745 return XDP_DROP;
746 data_stats->v1 += 1;
747 data_stats->v2 += pkt_bytes;
748
749 data = (void *)(long)xdp->data;
750 data_end = (void *)(long)xdp->data_end;
751 if (data + 4 > data_end)
752 return XDP_DROP;
753 *(u32 *)data = dst->dst;
754 return XDP_DROP;
755}
756
757SEC("xdp")
758int balancer_ingress_v4(struct xdp_md *ctx)
759{
760 void *data = (void *)(long)ctx->data;
761 void *data_end = (void *)(long)ctx->data_end;
762 struct eth_hdr *eth = data;
763 __u32 eth_proto;
764 __u32 nh_off;
765
766 nh_off = sizeof(struct eth_hdr);
767 if (data + nh_off > data_end)
768 return XDP_DROP;
769 eth_proto = bpf_ntohs(eth->eth_proto);
770 if (eth_proto == ETH_P_IP)
771 return process_packet(data, nh_off, data_end, 0, ctx);
772 else
773 return XDP_DROP;
774}
775
776SEC("xdp")
777int balancer_ingress_v6(struct xdp_md *ctx)
778{
779 void *data = (void *)(long)ctx->data;
780 void *data_end = (void *)(long)ctx->data_end;
781 struct eth_hdr *eth = data;
782 __u32 eth_proto;
783 __u32 nh_off;
784
785 nh_off = sizeof(struct eth_hdr);
786 if (data + nh_off > data_end)
787 return XDP_DROP;
788 eth_proto = bpf_ntohs(eth->eth_proto);
789 if (eth_proto == ETH_P_IPV6)
790 return process_packet(data, nh_off, data_end, 1, ctx);
791 else
792 return XDP_DROP;
793}
794
795char _license[] SEC("license") = "GPL";