xref: /linux/tools/testing/selftests/bpf/progs/test_l4lb.c (revision 96ac6d435100450f0565708d9b885ea2a7400e0a)
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_helpers.h"
21 #include "test_iptunnel_common.h"
22 #include "bpf_endian.h"
23 
24 int _version SEC("version") = 1;
25 
26 static inline __u32 rol32(__u32 word, unsigned int shift)
27 {
28 	return (word << shift) | (word >> ((-shift) & 31));
29 }
30 
31 /* copy paste of jhash from kernel sources to make sure llvm
32  * can compile it into valid sequence of bpf instructions
33  */
34 #define __jhash_mix(a, b, c)			\
35 {						\
36 	a -= c;  a ^= rol32(c, 4);  c += b;	\
37 	b -= a;  b ^= rol32(a, 6);  a += c;	\
38 	c -= b;  c ^= rol32(b, 8);  b += a;	\
39 	a -= c;  a ^= rol32(c, 16); c += b;	\
40 	b -= a;  b ^= rol32(a, 19); a += c;	\
41 	c -= b;  c ^= rol32(b, 4);  b += a;	\
42 }
43 
44 #define __jhash_final(a, b, c)			\
45 {						\
46 	c ^= b; c -= rol32(b, 14);		\
47 	a ^= c; a -= rol32(c, 11);		\
48 	b ^= a; b -= rol32(a, 25);		\
49 	c ^= b; c -= rol32(b, 16);		\
50 	a ^= c; a -= rol32(c, 4);		\
51 	b ^= a; b -= rol32(a, 14);		\
52 	c ^= b; c -= rol32(b, 24);		\
53 }
54 
55 #define JHASH_INITVAL		0xdeadbeef
56 
57 typedef unsigned int u32;
58 
59 static inline u32 jhash(const void *key, u32 length, u32 initval)
60 {
61 	u32 a, b, c;
62 	const unsigned char *k = key;
63 
64 	a = b = c = JHASH_INITVAL + length + initval;
65 
66 	while (length > 12) {
67 		a += *(u32 *)(k);
68 		b += *(u32 *)(k + 4);
69 		c += *(u32 *)(k + 8);
70 		__jhash_mix(a, b, c);
71 		length -= 12;
72 		k += 12;
73 	}
74 	switch (length) {
75 	case 12: c += (u32)k[11]<<24;
76 	case 11: c += (u32)k[10]<<16;
77 	case 10: c += (u32)k[9]<<8;
78 	case 9:  c += k[8];
79 	case 8:  b += (u32)k[7]<<24;
80 	case 7:  b += (u32)k[6]<<16;
81 	case 6:  b += (u32)k[5]<<8;
82 	case 5:  b += k[4];
83 	case 4:  a += (u32)k[3]<<24;
84 	case 3:  a += (u32)k[2]<<16;
85 	case 2:  a += (u32)k[1]<<8;
86 	case 1:  a += k[0];
87 		 __jhash_final(a, b, c);
88 	case 0: /* Nothing left to add */
89 		break;
90 	}
91 
92 	return c;
93 }
94 
95 static inline u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
96 {
97 	a += initval;
98 	b += initval;
99 	c += initval;
100 	__jhash_final(a, b, c);
101 	return c;
102 }
103 
104 static inline u32 jhash_2words(u32 a, u32 b, u32 initval)
105 {
106 	return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
107 }
108 
109 #define PCKT_FRAGMENTED 65343
110 #define IPV4_HDR_LEN_NO_OPT 20
111 #define IPV4_PLUS_ICMP_HDR 28
112 #define IPV6_PLUS_ICMP_HDR 48
113 #define RING_SIZE 2
114 #define MAX_VIPS 12
115 #define MAX_REALS 5
116 #define CTL_MAP_SIZE 16
117 #define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE)
118 #define F_IPV6 (1 << 0)
119 #define F_HASH_NO_SRC_PORT (1 << 0)
120 #define F_ICMP (1 << 0)
121 #define F_SYN_SET (1 << 1)
122 
123 struct packet_description {
124 	union {
125 		__be32 src;
126 		__be32 srcv6[4];
127 	};
128 	union {
129 		__be32 dst;
130 		__be32 dstv6[4];
131 	};
132 	union {
133 		__u32 ports;
134 		__u16 port16[2];
135 	};
136 	__u8 proto;
137 	__u8 flags;
138 };
139 
140 struct ctl_value {
141 	union {
142 		__u64 value;
143 		__u32 ifindex;
144 		__u8 mac[6];
145 	};
146 };
147 
148 struct vip_meta {
149 	__u32 flags;
150 	__u32 vip_num;
151 };
152 
153 struct real_definition {
154 	union {
155 		__be32 dst;
156 		__be32 dstv6[4];
157 	};
158 	__u8 flags;
159 };
160 
161 struct vip_stats {
162 	__u64 bytes;
163 	__u64 pkts;
164 };
165 
166 struct eth_hdr {
167 	unsigned char eth_dest[ETH_ALEN];
168 	unsigned char eth_source[ETH_ALEN];
169 	unsigned short eth_proto;
170 };
171 
172 struct bpf_map_def SEC("maps") vip_map = {
173 	.type = BPF_MAP_TYPE_HASH,
174 	.key_size = sizeof(struct vip),
175 	.value_size = sizeof(struct vip_meta),
176 	.max_entries = MAX_VIPS,
177 };
178 
179 struct bpf_map_def SEC("maps") ch_rings = {
180 	.type = BPF_MAP_TYPE_ARRAY,
181 	.key_size = sizeof(__u32),
182 	.value_size = sizeof(__u32),
183 	.max_entries = CH_RINGS_SIZE,
184 };
185 
186 struct bpf_map_def SEC("maps") reals = {
187 	.type = BPF_MAP_TYPE_ARRAY,
188 	.key_size = sizeof(__u32),
189 	.value_size = sizeof(struct real_definition),
190 	.max_entries = MAX_REALS,
191 };
192 
193 struct bpf_map_def SEC("maps") stats = {
194 	.type = BPF_MAP_TYPE_PERCPU_ARRAY,
195 	.key_size = sizeof(__u32),
196 	.value_size = sizeof(struct vip_stats),
197 	.max_entries = MAX_VIPS,
198 };
199 
200 struct bpf_map_def SEC("maps") ctl_array = {
201 	.type = BPF_MAP_TYPE_ARRAY,
202 	.key_size = sizeof(__u32),
203 	.value_size = sizeof(struct ctl_value),
204 	.max_entries = CTL_MAP_SIZE,
205 };
206 
207 static __always_inline __u32 get_packet_hash(struct packet_description *pckt,
208 					     bool ipv6)
209 {
210 	if (ipv6)
211 		return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
212 				    pckt->ports, CH_RINGS_SIZE);
213 	else
214 		return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
215 }
216 
217 static __always_inline bool get_packet_dst(struct real_definition **real,
218 					   struct packet_description *pckt,
219 					   struct vip_meta *vip_info,
220 					   bool is_ipv6)
221 {
222 	__u32 hash = get_packet_hash(pckt, is_ipv6) % RING_SIZE;
223 	__u32 key = RING_SIZE * vip_info->vip_num + hash;
224 	__u32 *real_pos;
225 
226 	real_pos = bpf_map_lookup_elem(&ch_rings, &key);
227 	if (!real_pos)
228 		return false;
229 	key = *real_pos;
230 	*real = bpf_map_lookup_elem(&reals, &key);
231 	if (!(*real))
232 		return false;
233 	return true;
234 }
235 
236 static __always_inline int parse_icmpv6(void *data, void *data_end, __u64 off,
237 					struct packet_description *pckt)
238 {
239 	struct icmp6hdr *icmp_hdr;
240 	struct ipv6hdr *ip6h;
241 
242 	icmp_hdr = data + off;
243 	if (icmp_hdr + 1 > data_end)
244 		return TC_ACT_SHOT;
245 	if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
246 		return TC_ACT_OK;
247 	off += sizeof(struct icmp6hdr);
248 	ip6h = data + off;
249 	if (ip6h + 1 > data_end)
250 		return TC_ACT_SHOT;
251 	pckt->proto = ip6h->nexthdr;
252 	pckt->flags |= F_ICMP;
253 	memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
254 	memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
255 	return TC_ACT_UNSPEC;
256 }
257 
258 static __always_inline int parse_icmp(void *data, void *data_end, __u64 off,
259 				      struct packet_description *pckt)
260 {
261 	struct icmphdr *icmp_hdr;
262 	struct iphdr *iph;
263 
264 	icmp_hdr = data + off;
265 	if (icmp_hdr + 1 > data_end)
266 		return TC_ACT_SHOT;
267 	if (icmp_hdr->type != ICMP_DEST_UNREACH ||
268 	    icmp_hdr->code != ICMP_FRAG_NEEDED)
269 		return TC_ACT_OK;
270 	off += sizeof(struct icmphdr);
271 	iph = data + off;
272 	if (iph + 1 > data_end)
273 		return TC_ACT_SHOT;
274 	if (iph->ihl != 5)
275 		return TC_ACT_SHOT;
276 	pckt->proto = iph->protocol;
277 	pckt->flags |= F_ICMP;
278 	pckt->src = iph->daddr;
279 	pckt->dst = iph->saddr;
280 	return TC_ACT_UNSPEC;
281 }
282 
283 static __always_inline bool parse_udp(void *data, __u64 off, void *data_end,
284 				      struct packet_description *pckt)
285 {
286 	struct udphdr *udp;
287 	udp = data + off;
288 
289 	if (udp + 1 > data_end)
290 		return false;
291 
292 	if (!(pckt->flags & F_ICMP)) {
293 		pckt->port16[0] = udp->source;
294 		pckt->port16[1] = udp->dest;
295 	} else {
296 		pckt->port16[0] = udp->dest;
297 		pckt->port16[1] = udp->source;
298 	}
299 	return true;
300 }
301 
302 static __always_inline bool parse_tcp(void *data, __u64 off, void *data_end,
303 				      struct packet_description *pckt)
304 {
305 	struct tcphdr *tcp;
306 
307 	tcp = data + off;
308 	if (tcp + 1 > data_end)
309 		return false;
310 
311 	if (tcp->syn)
312 		pckt->flags |= F_SYN_SET;
313 
314 	if (!(pckt->flags & F_ICMP)) {
315 		pckt->port16[0] = tcp->source;
316 		pckt->port16[1] = tcp->dest;
317 	} else {
318 		pckt->port16[0] = tcp->dest;
319 		pckt->port16[1] = tcp->source;
320 	}
321 	return true;
322 }
323 
324 static __always_inline int process_packet(void *data, __u64 off, void *data_end,
325 					  bool is_ipv6, struct __sk_buff *skb)
326 {
327 	void *pkt_start = (void *)(long)skb->data;
328 	struct packet_description pckt = {};
329 	struct eth_hdr *eth = pkt_start;
330 	struct bpf_tunnel_key tkey = {};
331 	struct vip_stats *data_stats;
332 	struct real_definition *dst;
333 	struct vip_meta *vip_info;
334 	struct ctl_value *cval;
335 	__u32 v4_intf_pos = 1;
336 	__u32 v6_intf_pos = 2;
337 	struct ipv6hdr *ip6h;
338 	struct vip vip = {};
339 	struct iphdr *iph;
340 	int tun_flag = 0;
341 	__u16 pkt_bytes;
342 	__u64 iph_len;
343 	__u32 ifindex;
344 	__u8 protocol;
345 	__u32 vip_num;
346 	int action;
347 
348 	tkey.tunnel_ttl = 64;
349 	if (is_ipv6) {
350 		ip6h = data + off;
351 		if (ip6h + 1 > data_end)
352 			return TC_ACT_SHOT;
353 
354 		iph_len = sizeof(struct ipv6hdr);
355 		protocol = ip6h->nexthdr;
356 		pckt.proto = protocol;
357 		pkt_bytes = bpf_ntohs(ip6h->payload_len);
358 		off += iph_len;
359 		if (protocol == IPPROTO_FRAGMENT) {
360 			return TC_ACT_SHOT;
361 		} else if (protocol == IPPROTO_ICMPV6) {
362 			action = parse_icmpv6(data, data_end, off, &pckt);
363 			if (action >= 0)
364 				return action;
365 			off += IPV6_PLUS_ICMP_HDR;
366 		} else {
367 			memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
368 			memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
369 		}
370 	} else {
371 		iph = data + off;
372 		if (iph + 1 > data_end)
373 			return TC_ACT_SHOT;
374 		if (iph->ihl != 5)
375 			return TC_ACT_SHOT;
376 
377 		protocol = iph->protocol;
378 		pckt.proto = protocol;
379 		pkt_bytes = bpf_ntohs(iph->tot_len);
380 		off += IPV4_HDR_LEN_NO_OPT;
381 
382 		if (iph->frag_off & PCKT_FRAGMENTED)
383 			return TC_ACT_SHOT;
384 		if (protocol == IPPROTO_ICMP) {
385 			action = parse_icmp(data, data_end, off, &pckt);
386 			if (action >= 0)
387 				return action;
388 			off += IPV4_PLUS_ICMP_HDR;
389 		} else {
390 			pckt.src = iph->saddr;
391 			pckt.dst = iph->daddr;
392 		}
393 	}
394 	protocol = pckt.proto;
395 
396 	if (protocol == IPPROTO_TCP) {
397 		if (!parse_tcp(data, off, data_end, &pckt))
398 			return TC_ACT_SHOT;
399 	} else if (protocol == IPPROTO_UDP) {
400 		if (!parse_udp(data, off, data_end, &pckt))
401 			return TC_ACT_SHOT;
402 	} else {
403 		return TC_ACT_SHOT;
404 	}
405 
406 	if (is_ipv6)
407 		memcpy(vip.daddr.v6, pckt.dstv6, 16);
408 	else
409 		vip.daddr.v4 = pckt.dst;
410 
411 	vip.dport = pckt.port16[1];
412 	vip.protocol = pckt.proto;
413 	vip_info = bpf_map_lookup_elem(&vip_map, &vip);
414 	if (!vip_info) {
415 		vip.dport = 0;
416 		vip_info = bpf_map_lookup_elem(&vip_map, &vip);
417 		if (!vip_info)
418 			return TC_ACT_SHOT;
419 		pckt.port16[1] = 0;
420 	}
421 
422 	if (vip_info->flags & F_HASH_NO_SRC_PORT)
423 		pckt.port16[0] = 0;
424 
425 	if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
426 		return TC_ACT_SHOT;
427 
428 	if (dst->flags & F_IPV6) {
429 		cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
430 		if (!cval)
431 			return TC_ACT_SHOT;
432 		ifindex = cval->ifindex;
433 		memcpy(tkey.remote_ipv6, dst->dstv6, 16);
434 		tun_flag = BPF_F_TUNINFO_IPV6;
435 	} else {
436 		cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
437 		if (!cval)
438 			return TC_ACT_SHOT;
439 		ifindex = cval->ifindex;
440 		tkey.remote_ipv4 = dst->dst;
441 	}
442 	vip_num = vip_info->vip_num;
443 	data_stats = bpf_map_lookup_elem(&stats, &vip_num);
444 	if (!data_stats)
445 		return TC_ACT_SHOT;
446 	data_stats->pkts++;
447 	data_stats->bytes += pkt_bytes;
448 	bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
449 	*(u32 *)eth->eth_dest = tkey.remote_ipv4;
450 	return bpf_redirect(ifindex, 0);
451 }
452 
453 SEC("l4lb-demo")
454 int balancer_ingress(struct __sk_buff *ctx)
455 {
456 	void *data_end = (void *)(long)ctx->data_end;
457 	void *data = (void *)(long)ctx->data;
458 	struct eth_hdr *eth = data;
459 	__u32 eth_proto;
460 	__u32 nh_off;
461 
462 	nh_off = sizeof(struct eth_hdr);
463 	if (data + nh_off > data_end)
464 		return TC_ACT_SHOT;
465 	eth_proto = eth->eth_proto;
466 	if (eth_proto == bpf_htons(ETH_P_IP))
467 		return process_packet(data, nh_off, data_end, false, ctx);
468 	else if (eth_proto == bpf_htons(ETH_P_IPV6))
469 		return process_packet(data, nh_off, data_end, true, ctx);
470 	else
471 		return TC_ACT_SHOT;
472 }
473 char _license[] SEC("license") = "GPL";
474