1 /* Copyright (c) 2017 Facebook
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of version 2 of the GNU General Public
5  * License as published by the Free Software Foundation.
6  */
7 #include <stddef.h>
8 #include <stdbool.h>
9 #include <string.h>
10 #include <linux/pkt_cls.h>
11 #include <linux/bpf.h>
12 #include <linux/in.h>
13 #include <linux/if_ether.h>
14 #include <linux/ip.h>
15 #include <linux/ipv6.h>
16 #include <linux/icmp.h>
17 #include <linux/icmpv6.h>
18 #include <linux/tcp.h>
19 #include <linux/udp.h>
20 #include <bpf/bpf_helpers.h>
21 #include "test_iptunnel_common.h"
22 #include <bpf/bpf_endian.h>
23 
24 static inline __u32 rol32(__u32 word, unsigned int shift)
25 {
26 	return (word << shift) | (word >> ((-shift) & 31));
27 }
28 
29 /* copy paste of jhash from kernel sources to make sure llvm
30  * can compile it into valid sequence of bpf instructions
31  */
32 #define __jhash_mix(a, b, c)			\
33 {						\
34 	a -= c;  a ^= rol32(c, 4);  c += b;	\
35 	b -= a;  b ^= rol32(a, 6);  a += c;	\
36 	c -= b;  c ^= rol32(b, 8);  b += a;	\
37 	a -= c;  a ^= rol32(c, 16); c += b;	\
38 	b -= a;  b ^= rol32(a, 19); a += c;	\
39 	c -= b;  c ^= rol32(b, 4);  b += a;	\
40 }
41 
42 #define __jhash_final(a, b, c)			\
43 {						\
44 	c ^= b; c -= rol32(b, 14);		\
45 	a ^= c; a -= rol32(c, 11);		\
46 	b ^= a; b -= rol32(a, 25);		\
47 	c ^= b; c -= rol32(b, 16);		\
48 	a ^= c; a -= rol32(c, 4);		\
49 	b ^= a; b -= rol32(a, 14);		\
50 	c ^= b; c -= rol32(b, 24);		\
51 }
52 
53 #define JHASH_INITVAL		0xdeadbeef
54 
55 typedef unsigned int u32;
56 
57 static inline u32 jhash(const void *key, u32 length, u32 initval)
58 {
59 	u32 a, b, c;
60 	const unsigned char *k = key;
61 
62 	a = b = c = JHASH_INITVAL + length + initval;
63 
64 	while (length > 12) {
65 		a += *(u32 *)(k);
66 		b += *(u32 *)(k + 4);
67 		c += *(u32 *)(k + 8);
68 		__jhash_mix(a, b, c);
69 		length -= 12;
70 		k += 12;
71 	}
72 	switch (length) {
73 	case 12: c += (u32)k[11]<<24;
74 	case 11: c += (u32)k[10]<<16;
75 	case 10: c += (u32)k[9]<<8;
76 	case 9:  c += k[8];
77 	case 8:  b += (u32)k[7]<<24;
78 	case 7:  b += (u32)k[6]<<16;
79 	case 6:  b += (u32)k[5]<<8;
80 	case 5:  b += k[4];
81 	case 4:  a += (u32)k[3]<<24;
82 	case 3:  a += (u32)k[2]<<16;
83 	case 2:  a += (u32)k[1]<<8;
84 	case 1:  a += k[0];
85 		 __jhash_final(a, b, c);
86 	case 0: /* Nothing left to add */
87 		break;
88 	}
89 
90 	return c;
91 }
92 
93 static inline u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
94 {
95 	a += initval;
96 	b += initval;
97 	c += initval;
98 	__jhash_final(a, b, c);
99 	return c;
100 }
101 
102 static inline u32 jhash_2words(u32 a, u32 b, u32 initval)
103 {
104 	return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
105 }
106 
107 #define PCKT_FRAGMENTED 65343
108 #define IPV4_HDR_LEN_NO_OPT 20
109 #define IPV4_PLUS_ICMP_HDR 28
110 #define IPV6_PLUS_ICMP_HDR 48
111 #define RING_SIZE 2
112 #define MAX_VIPS 12
113 #define MAX_REALS 5
114 #define CTL_MAP_SIZE 16
115 #define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE)
116 #define F_IPV6 (1 << 0)
117 #define F_HASH_NO_SRC_PORT (1 << 0)
118 #define F_ICMP (1 << 0)
119 #define F_SYN_SET (1 << 1)
120 
121 struct packet_description {
122 	union {
123 		__be32 src;
124 		__be32 srcv6[4];
125 	};
126 	union {
127 		__be32 dst;
128 		__be32 dstv6[4];
129 	};
130 	union {
131 		__u32 ports;
132 		__u16 port16[2];
133 	};
134 	__u8 proto;
135 	__u8 flags;
136 };
137 
138 struct ctl_value {
139 	union {
140 		__u64 value;
141 		__u32 ifindex;
142 		__u8 mac[6];
143 	};
144 };
145 
146 struct vip_meta {
147 	__u32 flags;
148 	__u32 vip_num;
149 };
150 
151 struct real_definition {
152 	union {
153 		__be32 dst;
154 		__be32 dstv6[4];
155 	};
156 	__u8 flags;
157 };
158 
159 struct vip_stats {
160 	__u64 bytes;
161 	__u64 pkts;
162 };
163 
164 struct eth_hdr {
165 	unsigned char eth_dest[ETH_ALEN];
166 	unsigned char eth_source[ETH_ALEN];
167 	unsigned short eth_proto;
168 };
169 
170 struct {
171 	__uint(type, BPF_MAP_TYPE_HASH);
172 	__uint(max_entries, MAX_VIPS);
173 	__type(key, struct vip);
174 	__type(value, struct vip_meta);
175 } vip_map SEC(".maps");
176 
177 struct {
178 	__uint(type, BPF_MAP_TYPE_ARRAY);
179 	__uint(max_entries, CH_RINGS_SIZE);
180 	__type(key, __u32);
181 	__type(value, __u32);
182 } ch_rings SEC(".maps");
183 
184 struct {
185 	__uint(type, BPF_MAP_TYPE_ARRAY);
186 	__uint(max_entries, MAX_REALS);
187 	__type(key, __u32);
188 	__type(value, struct real_definition);
189 } reals SEC(".maps");
190 
191 struct {
192 	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
193 	__uint(max_entries, MAX_VIPS);
194 	__type(key, __u32);
195 	__type(value, struct vip_stats);
196 } stats SEC(".maps");
197 
198 struct {
199 	__uint(type, BPF_MAP_TYPE_ARRAY);
200 	__uint(max_entries, CTL_MAP_SIZE);
201 	__type(key, __u32);
202 	__type(value, struct ctl_value);
203 } ctl_array SEC(".maps");
204 
205 static __always_inline __u32 get_packet_hash(struct packet_description *pckt,
206 					     bool ipv6)
207 {
208 	if (ipv6)
209 		return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
210 				    pckt->ports, CH_RINGS_SIZE);
211 	else
212 		return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
213 }
214 
215 static __always_inline bool get_packet_dst(struct real_definition **real,
216 					   struct packet_description *pckt,
217 					   struct vip_meta *vip_info,
218 					   bool is_ipv6)
219 {
220 	__u32 hash = get_packet_hash(pckt, is_ipv6) % RING_SIZE;
221 	__u32 key = RING_SIZE * vip_info->vip_num + hash;
222 	__u32 *real_pos;
223 
224 	real_pos = bpf_map_lookup_elem(&ch_rings, &key);
225 	if (!real_pos)
226 		return false;
227 	key = *real_pos;
228 	*real = bpf_map_lookup_elem(&reals, &key);
229 	if (!(*real))
230 		return false;
231 	return true;
232 }
233 
234 static __always_inline int parse_icmpv6(void *data, void *data_end, __u64 off,
235 					struct packet_description *pckt)
236 {
237 	struct icmp6hdr *icmp_hdr;
238 	struct ipv6hdr *ip6h;
239 
240 	icmp_hdr = data + off;
241 	if (icmp_hdr + 1 > data_end)
242 		return TC_ACT_SHOT;
243 	if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
244 		return TC_ACT_OK;
245 	off += sizeof(struct icmp6hdr);
246 	ip6h = data + off;
247 	if (ip6h + 1 > data_end)
248 		return TC_ACT_SHOT;
249 	pckt->proto = ip6h->nexthdr;
250 	pckt->flags |= F_ICMP;
251 	memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
252 	memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
253 	return TC_ACT_UNSPEC;
254 }
255 
256 static __always_inline int parse_icmp(void *data, void *data_end, __u64 off,
257 				      struct packet_description *pckt)
258 {
259 	struct icmphdr *icmp_hdr;
260 	struct iphdr *iph;
261 
262 	icmp_hdr = data + off;
263 	if (icmp_hdr + 1 > data_end)
264 		return TC_ACT_SHOT;
265 	if (icmp_hdr->type != ICMP_DEST_UNREACH ||
266 	    icmp_hdr->code != ICMP_FRAG_NEEDED)
267 		return TC_ACT_OK;
268 	off += sizeof(struct icmphdr);
269 	iph = data + off;
270 	if (iph + 1 > data_end)
271 		return TC_ACT_SHOT;
272 	if (iph->ihl != 5)
273 		return TC_ACT_SHOT;
274 	pckt->proto = iph->protocol;
275 	pckt->flags |= F_ICMP;
276 	pckt->src = iph->daddr;
277 	pckt->dst = iph->saddr;
278 	return TC_ACT_UNSPEC;
279 }
280 
281 static __always_inline bool parse_udp(void *data, __u64 off, void *data_end,
282 				      struct packet_description *pckt)
283 {
284 	struct udphdr *udp;
285 	udp = data + off;
286 
287 	if (udp + 1 > data_end)
288 		return false;
289 
290 	if (!(pckt->flags & F_ICMP)) {
291 		pckt->port16[0] = udp->source;
292 		pckt->port16[1] = udp->dest;
293 	} else {
294 		pckt->port16[0] = udp->dest;
295 		pckt->port16[1] = udp->source;
296 	}
297 	return true;
298 }
299 
300 static __always_inline bool parse_tcp(void *data, __u64 off, void *data_end,
301 				      struct packet_description *pckt)
302 {
303 	struct tcphdr *tcp;
304 
305 	tcp = data + off;
306 	if (tcp + 1 > data_end)
307 		return false;
308 
309 	if (tcp->syn)
310 		pckt->flags |= F_SYN_SET;
311 
312 	if (!(pckt->flags & F_ICMP)) {
313 		pckt->port16[0] = tcp->source;
314 		pckt->port16[1] = tcp->dest;
315 	} else {
316 		pckt->port16[0] = tcp->dest;
317 		pckt->port16[1] = tcp->source;
318 	}
319 	return true;
320 }
321 
322 static __always_inline int process_packet(void *data, __u64 off, void *data_end,
323 					  bool is_ipv6, struct __sk_buff *skb)
324 {
325 	void *pkt_start = (void *)(long)skb->data;
326 	struct packet_description pckt = {};
327 	struct eth_hdr *eth = pkt_start;
328 	struct bpf_tunnel_key tkey = {};
329 	struct vip_stats *data_stats;
330 	struct real_definition *dst;
331 	struct vip_meta *vip_info;
332 	struct ctl_value *cval;
333 	__u32 v4_intf_pos = 1;
334 	__u32 v6_intf_pos = 2;
335 	struct ipv6hdr *ip6h;
336 	struct vip vip = {};
337 	struct iphdr *iph;
338 	int tun_flag = 0;
339 	__u16 pkt_bytes;
340 	__u64 iph_len;
341 	__u32 ifindex;
342 	__u8 protocol;
343 	__u32 vip_num;
344 	int action;
345 
346 	tkey.tunnel_ttl = 64;
347 	if (is_ipv6) {
348 		ip6h = data + off;
349 		if (ip6h + 1 > data_end)
350 			return TC_ACT_SHOT;
351 
352 		iph_len = sizeof(struct ipv6hdr);
353 		protocol = ip6h->nexthdr;
354 		pckt.proto = protocol;
355 		pkt_bytes = bpf_ntohs(ip6h->payload_len);
356 		off += iph_len;
357 		if (protocol == IPPROTO_FRAGMENT) {
358 			return TC_ACT_SHOT;
359 		} else if (protocol == IPPROTO_ICMPV6) {
360 			action = parse_icmpv6(data, data_end, off, &pckt);
361 			if (action >= 0)
362 				return action;
363 			off += IPV6_PLUS_ICMP_HDR;
364 		} else {
365 			memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
366 			memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
367 		}
368 	} else {
369 		iph = data + off;
370 		if (iph + 1 > data_end)
371 			return TC_ACT_SHOT;
372 		if (iph->ihl != 5)
373 			return TC_ACT_SHOT;
374 
375 		protocol = iph->protocol;
376 		pckt.proto = protocol;
377 		pkt_bytes = bpf_ntohs(iph->tot_len);
378 		off += IPV4_HDR_LEN_NO_OPT;
379 
380 		if (iph->frag_off & PCKT_FRAGMENTED)
381 			return TC_ACT_SHOT;
382 		if (protocol == IPPROTO_ICMP) {
383 			action = parse_icmp(data, data_end, off, &pckt);
384 			if (action >= 0)
385 				return action;
386 			off += IPV4_PLUS_ICMP_HDR;
387 		} else {
388 			pckt.src = iph->saddr;
389 			pckt.dst = iph->daddr;
390 		}
391 	}
392 	protocol = pckt.proto;
393 
394 	if (protocol == IPPROTO_TCP) {
395 		if (!parse_tcp(data, off, data_end, &pckt))
396 			return TC_ACT_SHOT;
397 	} else if (protocol == IPPROTO_UDP) {
398 		if (!parse_udp(data, off, data_end, &pckt))
399 			return TC_ACT_SHOT;
400 	} else {
401 		return TC_ACT_SHOT;
402 	}
403 
404 	if (is_ipv6)
405 		memcpy(vip.daddr.v6, pckt.dstv6, 16);
406 	else
407 		vip.daddr.v4 = pckt.dst;
408 
409 	vip.dport = pckt.port16[1];
410 	vip.protocol = pckt.proto;
411 	vip_info = bpf_map_lookup_elem(&vip_map, &vip);
412 	if (!vip_info) {
413 		vip.dport = 0;
414 		vip_info = bpf_map_lookup_elem(&vip_map, &vip);
415 		if (!vip_info)
416 			return TC_ACT_SHOT;
417 		pckt.port16[1] = 0;
418 	}
419 
420 	if (vip_info->flags & F_HASH_NO_SRC_PORT)
421 		pckt.port16[0] = 0;
422 
423 	if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
424 		return TC_ACT_SHOT;
425 
426 	if (dst->flags & F_IPV6) {
427 		cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
428 		if (!cval)
429 			return TC_ACT_SHOT;
430 		ifindex = cval->ifindex;
431 		memcpy(tkey.remote_ipv6, dst->dstv6, 16);
432 		tun_flag = BPF_F_TUNINFO_IPV6;
433 	} else {
434 		cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
435 		if (!cval)
436 			return TC_ACT_SHOT;
437 		ifindex = cval->ifindex;
438 		tkey.remote_ipv4 = dst->dst;
439 	}
440 	vip_num = vip_info->vip_num;
441 	data_stats = bpf_map_lookup_elem(&stats, &vip_num);
442 	if (!data_stats)
443 		return TC_ACT_SHOT;
444 	data_stats->pkts++;
445 	data_stats->bytes += pkt_bytes;
446 	bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
447 	*(u32 *)eth->eth_dest = tkey.remote_ipv4;
448 	return bpf_redirect(ifindex, 0);
449 }
450 
451 SEC("tc")
452 int balancer_ingress(struct __sk_buff *ctx)
453 {
454 	void *data_end = (void *)(long)ctx->data_end;
455 	void *data = (void *)(long)ctx->data;
456 	struct eth_hdr *eth = data;
457 	__u32 eth_proto;
458 	__u32 nh_off;
459 
460 	nh_off = sizeof(struct eth_hdr);
461 	if (data + nh_off > data_end)
462 		return TC_ACT_SHOT;
463 	eth_proto = eth->eth_proto;
464 	if (eth_proto == bpf_htons(ETH_P_IP))
465 		return process_packet(data, nh_off, data_end, false, ctx);
466 	else if (eth_proto == bpf_htons(ETH_P_IPV6))
467 		return process_packet(data, nh_off, data_end, true, ctx);
468 	else
469 		return TC_ACT_SHOT;
470 }
471 char _license[] SEC("license") = "GPL";
472