1// SPDX-License-Identifier: GPL-2.0
2// Copyright (c) 2017 Facebook
3#include <stddef.h>
4#include <stdbool.h>
5#include <string.h>
6#include <linux/pkt_cls.h>
7#include <linux/bpf.h>
8#include <linux/in.h>
9#include <linux/if_ether.h>
10#include <linux/ip.h>
11#include <linux/ipv6.h>
12#include <linux/icmp.h>
13#include <linux/icmpv6.h>
14#include <linux/tcp.h>
15#include <linux/udp.h>
16#include <bpf/bpf_helpers.h>
17#include "test_iptunnel_common.h"
18#include <bpf/bpf_endian.h>
19
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 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
89static __noinline u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval)
90{
91	a += initval;
92	b += initval;
93	c += initval;
94	__jhash_final(a, b, c);
95	return c;
96}
97
98static __noinline u32 jhash_2words(u32 a, u32 b, u32 initval)
99{
100	return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2));
101}
102
103#define PCKT_FRAGMENTED 65343
104#define IPV4_HDR_LEN_NO_OPT 20
105#define IPV4_PLUS_ICMP_HDR 28
106#define IPV6_PLUS_ICMP_HDR 48
107#define RING_SIZE 2
108#define MAX_VIPS 12
109#define MAX_REALS 5
110#define CTL_MAP_SIZE 16
111#define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE)
112#define F_IPV6 (1 << 0)
113#define F_HASH_NO_SRC_PORT (1 << 0)
114#define F_ICMP (1 << 0)
115#define F_SYN_SET (1 << 1)
116
117struct packet_description {
118	union {
119		__be32 src;
120		__be32 srcv6[4];
121	};
122	union {
123		__be32 dst;
124		__be32 dstv6[4];
125	};
126	union {
127		__u32 ports;
128		__u16 port16[2];
129	};
130	__u8 proto;
131	__u8 flags;
132};
133
134struct ctl_value {
135	union {
136		__u64 value;
137		__u32 ifindex;
138		__u8 mac[6];
139	};
140};
141
142struct vip_meta {
143	__u32 flags;
144	__u32 vip_num;
145};
146
147struct real_definition {
148	union {
149		__be32 dst;
150		__be32 dstv6[4];
151	};
152	__u8 flags;
153};
154
155struct vip_stats {
156	__u64 bytes;
157	__u64 pkts;
158};
159
160struct eth_hdr {
161	unsigned char eth_dest[ETH_ALEN];
162	unsigned char eth_source[ETH_ALEN];
163	unsigned short eth_proto;
164};
165
166struct {
167	__uint(type, BPF_MAP_TYPE_HASH);
168	__uint(max_entries, MAX_VIPS);
169	__type(key, struct vip);
170	__type(value, struct vip_meta);
171} vip_map SEC(".maps");
172
173struct {
174	__uint(type, BPF_MAP_TYPE_ARRAY);
175	__uint(max_entries, CH_RINGS_SIZE);
176	__type(key, __u32);
177	__type(value, __u32);
178} ch_rings SEC(".maps");
179
180struct {
181	__uint(type, BPF_MAP_TYPE_ARRAY);
182	__uint(max_entries, MAX_REALS);
183	__type(key, __u32);
184	__type(value, struct real_definition);
185} reals SEC(".maps");
186
187struct {
188	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
189	__uint(max_entries, MAX_VIPS);
190	__type(key, __u32);
191	__type(value, struct vip_stats);
192} stats SEC(".maps");
193
194struct {
195	__uint(type, BPF_MAP_TYPE_ARRAY);
196	__uint(max_entries, CTL_MAP_SIZE);
197	__type(key, __u32);
198	__type(value, struct ctl_value);
199} ctl_array SEC(".maps");
200
201static __noinline __u32 get_packet_hash(struct packet_description *pckt, bool ipv6)
202{
203	if (ipv6)
204		return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS),
205				    pckt->ports, CH_RINGS_SIZE);
206	else
207		return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE);
208}
209
210static __noinline bool get_packet_dst(struct real_definition **real,
211				      struct packet_description *pckt,
212				      struct vip_meta *vip_info,
213				      bool is_ipv6)
214{
215	__u32 hash = get_packet_hash(pckt, is_ipv6);
216	__u32 key = RING_SIZE * vip_info->vip_num + hash % RING_SIZE;
217	__u32 *real_pos;
218
219	if (hash != 0x358459b7 /* jhash of ipv4 packet */  &&
220	    hash != 0x2f4bc6bb /* jhash of ipv6 packet */)
221		return false;
222
223	real_pos = bpf_map_lookup_elem(&ch_rings, &key);
224	if (!real_pos)
225		return false;
226	key = *real_pos;
227	*real = bpf_map_lookup_elem(&reals, &key);
228	if (!(*real))
229		return false;
230	return true;
231}
232
233static __noinline int parse_icmpv6(void *data, void *data_end, __u64 off,
234				   struct packet_description *pckt)
235{
236	struct icmp6hdr *icmp_hdr;
237	struct ipv6hdr *ip6h;
238
239	icmp_hdr = data + off;
240	if (icmp_hdr + 1 > data_end)
241		return TC_ACT_SHOT;
242	if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG)
243		return TC_ACT_OK;
244	off += sizeof(struct icmp6hdr);
245	ip6h = data + off;
246	if (ip6h + 1 > data_end)
247		return TC_ACT_SHOT;
248	pckt->proto = ip6h->nexthdr;
249	pckt->flags |= F_ICMP;
250	memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16);
251	memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16);
252	return TC_ACT_UNSPEC;
253}
254
255static __noinline int parse_icmp(void *data, void *data_end, __u64 off,
256				 struct packet_description *pckt)
257{
258	struct icmphdr *icmp_hdr;
259	struct iphdr *iph;
260
261	icmp_hdr = data + off;
262	if (icmp_hdr + 1 > data_end)
263		return TC_ACT_SHOT;
264	if (icmp_hdr->type != ICMP_DEST_UNREACH ||
265	    icmp_hdr->code != ICMP_FRAG_NEEDED)
266		return TC_ACT_OK;
267	off += sizeof(struct icmphdr);
268	iph = data + off;
269	if (iph + 1 > data_end)
270		return TC_ACT_SHOT;
271	if (iph->ihl != 5)
272		return TC_ACT_SHOT;
273	pckt->proto = iph->protocol;
274	pckt->flags |= F_ICMP;
275	pckt->src = iph->daddr;
276	pckt->dst = iph->saddr;
277	return TC_ACT_UNSPEC;
278}
279
280static __noinline bool parse_udp(void *data, __u64 off, void *data_end,
281				 struct packet_description *pckt)
282{
283	struct udphdr *udp;
284	udp = data + off;
285
286	if (udp + 1 > data_end)
287		return false;
288
289	if (!(pckt->flags & F_ICMP)) {
290		pckt->port16[0] = udp->source;
291		pckt->port16[1] = udp->dest;
292	} else {
293		pckt->port16[0] = udp->dest;
294		pckt->port16[1] = udp->source;
295	}
296	return true;
297}
298
299static __noinline bool parse_tcp(void *data, __u64 off, void *data_end,
300				 struct packet_description *pckt)
301{
302	struct tcphdr *tcp;
303
304	tcp = data + off;
305	if (tcp + 1 > data_end)
306		return false;
307
308	if (tcp->syn)
309		pckt->flags |= F_SYN_SET;
310
311	if (!(pckt->flags & F_ICMP)) {
312		pckt->port16[0] = tcp->source;
313		pckt->port16[1] = tcp->dest;
314	} else {
315		pckt->port16[0] = tcp->dest;
316		pckt->port16[1] = tcp->source;
317	}
318	return true;
319}
320
321static __noinline int process_packet(void *data, __u64 off, void *data_end,
322				     bool is_ipv6, struct __sk_buff *skb)
323{
324	void *pkt_start = (void *)(long)skb->data;
325	struct packet_description pckt = {};
326	struct eth_hdr *eth = pkt_start;
327	struct bpf_tunnel_key tkey = {};
328	struct vip_stats *data_stats;
329	struct real_definition *dst;
330	struct vip_meta *vip_info;
331	struct ctl_value *cval;
332	__u32 v4_intf_pos = 1;
333	__u32 v6_intf_pos = 2;
334	struct ipv6hdr *ip6h;
335	struct vip vip = {};
336	struct iphdr *iph;
337	int tun_flag = 0;
338	__u16 pkt_bytes;
339	__u64 iph_len;
340	__u32 ifindex;
341	__u8 protocol;
342	__u32 vip_num;
343	int action;
344
345	tkey.tunnel_ttl = 64;
346	if (is_ipv6) {
347		ip6h = data + off;
348		if (ip6h + 1 > data_end)
349			return TC_ACT_SHOT;
350
351		iph_len = sizeof(struct ipv6hdr);
352		protocol = ip6h->nexthdr;
353		pckt.proto = protocol;
354		pkt_bytes = bpf_ntohs(ip6h->payload_len);
355		off += iph_len;
356		if (protocol == IPPROTO_FRAGMENT) {
357			return TC_ACT_SHOT;
358		} else if (protocol == IPPROTO_ICMPV6) {
359			action = parse_icmpv6(data, data_end, off, &pckt);
360			if (action >= 0)
361				return action;
362			off += IPV6_PLUS_ICMP_HDR;
363		} else {
364			memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16);
365			memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16);
366		}
367	} else {
368		iph = data + off;
369		if (iph + 1 > data_end)
370			return TC_ACT_SHOT;
371		if (iph->ihl != 5)
372			return TC_ACT_SHOT;
373
374		protocol = iph->protocol;
375		pckt.proto = protocol;
376		pkt_bytes = bpf_ntohs(iph->tot_len);
377		off += IPV4_HDR_LEN_NO_OPT;
378
379		if (iph->frag_off & PCKT_FRAGMENTED)
380			return TC_ACT_SHOT;
381		if (protocol == IPPROTO_ICMP) {
382			action = parse_icmp(data, data_end, off, &pckt);
383			if (action >= 0)
384				return action;
385			off += IPV4_PLUS_ICMP_HDR;
386		} else {
387			pckt.src = iph->saddr;
388			pckt.dst = iph->daddr;
389		}
390	}
391	protocol = pckt.proto;
392
393	if (protocol == IPPROTO_TCP) {
394		if (!parse_tcp(data, off, data_end, &pckt))
395			return TC_ACT_SHOT;
396	} else if (protocol == IPPROTO_UDP) {
397		if (!parse_udp(data, off, data_end, &pckt))
398			return TC_ACT_SHOT;
399	} else {
400		return TC_ACT_SHOT;
401	}
402
403	if (is_ipv6)
404		memcpy(vip.daddr.v6, pckt.dstv6, 16);
405	else
406		vip.daddr.v4 = pckt.dst;
407
408	vip.dport = pckt.port16[1];
409	vip.protocol = pckt.proto;
410	vip_info = bpf_map_lookup_elem(&vip_map, &vip);
411	if (!vip_info) {
412		vip.dport = 0;
413		vip_info = bpf_map_lookup_elem(&vip_map, &vip);
414		if (!vip_info)
415			return TC_ACT_SHOT;
416		pckt.port16[1] = 0;
417	}
418
419	if (vip_info->flags & F_HASH_NO_SRC_PORT)
420		pckt.port16[0] = 0;
421
422	if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6))
423		return TC_ACT_SHOT;
424
425	if (dst->flags & F_IPV6) {
426		cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos);
427		if (!cval)
428			return TC_ACT_SHOT;
429		ifindex = cval->ifindex;
430		memcpy(tkey.remote_ipv6, dst->dstv6, 16);
431		tun_flag = BPF_F_TUNINFO_IPV6;
432	} else {
433		cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos);
434		if (!cval)
435			return TC_ACT_SHOT;
436		ifindex = cval->ifindex;
437		tkey.remote_ipv4 = dst->dst;
438	}
439	vip_num = vip_info->vip_num;
440	data_stats = bpf_map_lookup_elem(&stats, &vip_num);
441	if (!data_stats)
442		return TC_ACT_SHOT;
443	data_stats->pkts++;
444	data_stats->bytes += pkt_bytes;
445	bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag);
446	*(u32 *)eth->eth_dest = tkey.remote_ipv4;
447	return bpf_redirect(ifindex, 0);
448}
449
450SEC("tc")
451int balancer_ingress(struct __sk_buff *ctx)
452{
453	void *data_end = (void *)(long)ctx->data_end;
454	void *data = (void *)(long)ctx->data;
455	struct eth_hdr *eth = data;
456	__u32 eth_proto;
457	__u32 nh_off;
458
459	nh_off = sizeof(struct eth_hdr);
460	if (data + nh_off > data_end)
461		return TC_ACT_SHOT;
462	eth_proto = eth->eth_proto;
463	if (eth_proto == bpf_htons(ETH_P_IP))
464		return process_packet(data, nh_off, data_end, false, ctx);
465	else if (eth_proto == bpf_htons(ETH_P_IPV6))
466		return process_packet(data, nh_off, data_end, true, ctx);
467	else
468		return TC_ACT_SHOT;
469}
470char _license[] SEC("license") = "GPL";
471